# The AI Agenda: GPT5 leaks and the business of AI News — Steph Palazzolo, The Information Latent Space · 2025-08-06 Steph Palazzolo, AI journalist at The Information, explains how she covers the secretive AI industry, from OpenAI's GPT-5 to the inference market where startups like Modal raise billions. She reveals that inference providers are essentially GPU resellers facing margin pressure, and that GPT-5's performance will signal whether pre-training scaling or reinforcement learning drives progress. She critiques Meta's super intelligence ambition as misaligned with its ad-revenue business, and details the talent war with rumored hundred-million-dollar offers from Meta to poach researchers. Palazzolo also discusses the Windsurf acqui-hire backlash, the rise of coding agents like Claude Code, and how journalists protect sources while navigating leaks from CEOs like Sam Altman. ## Questions this episode answers ### How does AI journalist Steph Palazzolo get her scoops for The Information? Steph spends most of her day on calls with founders, researchers, and investors, asking for gossip and trends. She cold calls aggressively, builds long‑term trust with sources (often talking off‑the‑record first), and verifies information through multiple sources. She also scrutinizes each source’s incentives to avoid being manipulated by self‑serving leaks. [51:58](https://addtry.com/8339d7b7-73bc-4527-b041-140d3953dac5?t=3118000) ### What are the early impressions of GPT‑5 according to The Information’s sources? Sources who tested GPT‑5 told Steph Palazzolo they were impressed, with improvements in science, coding, creative writing, and general knowledge. The bigger question it raises is whether the gains come from continued pre‑training scaling or from advances in reinforcement learning—a distinction that will shape expectations for future models like o4. [26:57](https://addtry.com/8339d7b7-73bc-4527-b041-140d3953dac5?t=1617000) ### How do reporters verify the rumored billion‑dollar AI talent offers? Steph Palazzolo explains that journalists must consider a source’s incentives and confirm numbers with multiple sources. They dig into details like how much is cash versus stock, over how many years, because billion‑dollar figures can be manipulated. She notes that CEOs sometimes throw out eye‑catching numbers even without underlying verification. [20:56](https://addtry.com/8339d7b7-73bc-4527-b041-140d3953dac5?t=1256000) ## Key moments - **[0:00] Intro** - [1:11] Steph Palazzolo left investment banking to cover AI at The Information and write the AI Agenda newsletter. - [2:22] Steph Palazzolo's first generative AI story in summer 2022 covered VCs' obsession with DALL-E and generative AI. - **[3:23] The Beat** - [3:48] The Information has about 10 writers covering different areas of AI, says Steph Palazzolo. - [4:19] The Information's enterprise tech team includes reporters covering Google, Amazon, Microsoft, cloud, chips, and crypto. - [6:18] Steph Palazzolo spends only 20% of her day writing; the rest is calling sources for AI gossip and trends. - [7:31] Steph Palazzolo focuses on AI business models, such as inference providers' gross margins and financials. - **[7:55] Inference Boom** - [7:55] Modal raised $1 billion and Fireworks raised at a $4 billion valuation, heating up the inference market. - [8:13] Inference providers like Modal and Fireworks face skepticism as GPU resellers, but investors see explosive inference demand. - [10:53] Alessio: VCs overestimated how quickly existing cloud providers would catch up in inference, similar to the coding space. - [12:06] Fal.ai raised $125 million today, reflecting the capital-intensive inference market. - [13:01] Swyx compares inference providers to banks, doing maturity transformation of GPU contracts. - [13:16] Inference providers' success is a proxy for the gap between open and closed AI models, says Swyx. - **[15:07] Meta's Bet** - [15:18] Steph Palazzolo's newsletter argued Meta should not pursue super intelligence. - [15:51] Meta makes money from social media and ads, not super intelligence; Palazzolo urges focus on product innovation over huge models. - [19:23] Meta's 'personal super intelligence' marketing contrasts with Anthropic and OpenAI's focus on world-changing science. - [20:05] Meta spokesperson pushed back on rumored billion-dollar offers for AI researchers, reports Swyx. - [20:56] Steph Palazzolo verifies scoops about AI talent offers by checking multiple sources, deal structures, and incentives. - [23:32] Sam Altman claimed on a podcast that Mark Zuckerberg gave $100 million offers to OpenAI researchers, which became an immediate headline. - **[26:38] GPT-5** - [26:50] The Information is hiring a dedicated OpenAI writer, reflecting the company's complexity and newsworthiness. - [27:57] Early testers of GPT-5 are impressed by improvements in science, coding, creative writing, and general knowledge, reports Palazzolo. - [29:10] GPT-5's performance will reveal whether AI progress relies on pre-training scaling or reinforcement learning, says Palazzolo. - [30:28] GPT-5 is the main model on the horizon, with other labs likely a couple months behind, says Palazzolo. - [32:23] Anthropic's leaked memo about Middle East funding was an obvious signal of fundraising, says Palazzolo. - **[35:20] Coding War** - [35:53] Anthropic chose coding as its battleground, and developers like Claude for practical tasks beyond academic benchmarks. - [36:59] Anthropic's $200/month unlimited Claude Code plan helps adoption and provides training data, says Alessio. - [39:15] Swyx predicts AI coding assistants will eventually cost thousands per month as they prove their productivity value. - [40:02] Anthropic's revenue projections jumped from $4B to $9B due to Claude Code, but margins are a concern. - [40:18] OpenAI reached $12B ARR but projected burn rose from $1B to $8B, highlighting margin concerns. - [41:27] Claude Code leads briefly joined Cursor but returned to Anthropic after two weeks; comp wasn't the only factor. - **[43:15] Windsurf** - [43:28] Windsurf's acqui-hire by Cognition drew more backlash than similar deals because many employees felt left behind. - [46:02] VCs are growing tired of acqui-hire deals in AI; they invest for IPOs, not disappointing exits. - [47:42] Palazzolo covers coding agents by relying on developer sources and noting tools like Cline and Lovable. - [49:25] Swyx says legacy tech media like Forbes and Business Insider have declined, while The Information remains a top source. - **[49:35] Media Gap** - [49:57] The Information charges $500 subscriptions because ad models don't work, but Palazzolo sees media losing trust to influencers. - **[51:58] Scoop Craft** - [52:13] Palazzolo gets scoops by being curious and disarming, not aggressive; she cold-calls sources relentlessly. - [55:25] Palazzolo protects sources by withholding stories if needed and using vague attributions like 'person with knowledge'. - [57:19] The Information's AI stories are straightforward and try to explain complex topics like model weights in layman's terms. - [59:52] Palazzolo tracks AI trends by keeping a notepad of top five priorities and weekly editor meetings. - **[1:02:00] Pitching Press** - [1:02:57] Palazzolo: To get coverage, startups should pitch the importance of their entire space, not just their company. - **[1:05:36] Next Wave** - [1:06:03] Palazzolo is most interested in what comes after reasoning models: RL environments for agents, alternative architectures? - [1:08:13] Zuckerberg: 'On average today, people have like four friends, but we really have demand for like 12.' - [1:11:54] Applied Compute, a RL environments startup, is hyped as 'three kids in a basement worth $500 million.' - [1:12:12] Palazzolo wrote about RL environments where AI agents learn by roaming simulated apps like 'escape rooms'. ## Speakers - **Alessio** (host) - **Swyx** (host) - **Steph Palazzolo** (guest) ## Topics Inference, Language Models, Hiring ## Mentioned Anthropic (company), Cognition (company), DeepSeek (company), Fireworks (company), Lovable (company), Meta (company), Modal (company), OpenAI (company), The Information (company), Together (company), Bull (product), ChatGPT (product), Claude Code (product), Cline (product), Codex (product), Cursor (product), Devin (product), GPT (product), Llama (product), Windsurf (product) ## Transcript ### Intro **Alessio** [0:04] Everyone, welcome back to the Leading Space Podcast. This is Alessio, partner and CTO at Decibel, and I'm joined by Swyx, founder of Smile AI. **Swyx** [0:11] Hello, hello. Today we're... It's an unusual podcast because we're in a remote studio, but we're joined not by researcher or founder, but a journalist, our, our first-ever journalist on the podcast. Welcome Steph Palazzolo from The Information. **Steph Palazzolo** [0:24] Hi. Yeah, I'm super happy to be here. **Swyx** [0:27] Steph, I don't know if you remember, but actually, I think the first time I came across your work, I actually did a parody of it for the Sam Altman blip back in, like, 2023, where I, I put, like, sort of Adam Neumann as, like, the pot- prospective new CEO of OpenAI. **Steph Palazzolo** [0:42] Yeah. **Swyx** [0:42] That was pretty fun. **Steph Palazzolo** [0:43] No, that was super funny. I remember just seeing, like, a bunch of parodies of that over the next couple days. It was just, like, every single insane tech personality you could think of. Um, so yeah, that was a very insane week, but that, that meme definitely stands out to me. I definitely remember that. **Swyx** [0:59] I was like, "Who's the funniest possible CEO I could install in OpenAI? I know, the, the WeWork guy." But yeah, may-may... perhaps you could introduce yourself to our audience, uh, how'd you get started in, uh, reporting and basically covering AI for The Information. **Steph Palazzolo** [1:11] Totally, yeah. So, um, my name's Stephanie Palazzolo. I cover AI here at The Information, and so that means, um, working on just stories about the space, whether that's about, um, investments or big tech companies, broader trends. But I also help to run our daily AI newsletter, um, AI Agenda. So that comes out four times a week, and we get more into, like, the weeds and talk a bit more about what we're seeing with developers and, and researchers. Um, and so I've been here for about two years. Before that, I was covering the same beat at Business Insider. And then before that, actually, I ha- I have a bit of a weird background, because I used to work in tech investment banking, so I was, like, in finance, like, doing the IPO and any craziness during, uh, COVID. And yeah, I really enjoyed working with tech, um, tech companies and startups, but I was like, "I don't know if finance is my passion," which I don't think is, like, the craziest thing for anyone to say. But, um, I just had always loved journalism growing up and I was like, what better time in my life to try something kind of out there and try to combine, like, tech and writing and see how it goes. And yeah, I mean, everything's worked out, so I'm very happy. **Swyx** [2:16] Yeah, in, in, in remarkable fashion. Yeah, Alessio, you wanted to say something. **Alessio** [2:20] Uh, what was your first story? **Steph Palazzolo** [2:22] Yeah. **Alessio** [2:22] How did, how, how did you get started covering? **Steph Palazzolo** [2:25] Yeah. I mean, I think... I mean, I'm sure that my first story was probably just, like, some silly, like, write-up of, like, a report or something. But I think the first story I remember doing about AI was during summer 2022. I, I live in New York, so every couple months I go visit SF. And specifically on that trip I went and everyone was like, "Oh my God, let me show you this, like, funny app called DALL-E," and like, "We're gonna make pictures of, like, cats floating in space," and it's, like, so funny and cool. And then it just started to c- to come up in so many conversations that I was like, huh, like, this is, like, a fun little trend. Like, maybe I should write something about it. So then I ended up doing a piece that was like, why VCs are obsessed with, like, this new area called generative AI. And then obviously, like, three months later it was, like, blown up and ChatGPT was, was released. Um, but yeah. I remember that was my, like, first generative AI specific piece that I did, which was, like, honestly kind of just like an accident. Like, I just happened to talk with a bunch of VCs who were excited about it. Um, but yeah. And then ever since then I've been, been covering the space, so. **Swyx** [3:23] Yeah. Slowly f- being pulled into the orbit. Before we go deeper into y- your work and how you cover AI, I just wanted to k- discuss The Information more, more broadly. Um, I actually remember when Jessica Lessin left, like, uh, I think Wall- Wall Street Journal, um, and, uh, joined and, and started The Information. ### The Beat **Swyx** [3:39] I, I was basically following her since then. Um, and the, basically I'm just kinda... Yeah, like, I- I'm also a finance refugee, by the way. I, I used to be in a hedge fund as well. **Steph Palazzolo** [3:48] Nice. Yeah, nice. **Swyx** [3:49] And like, uh- **Steph Palazzolo** [3:50] All, all, uh, recovering ex-finance people. Love that. **Swyx** [3:53] Yeah. I mean, like, look, I, I think we have an interesting perspective in terms of, like, covering these startups as effectively stocks. Like, you know, I, I view myself secretly as, like, you know, all, all the write-ups that I do are industry sector reports that analysts may cover, except they're not publicly listed companies. So, like, uh, I'm, I'm curious, like, how is The Information organized today? AI has taken over everything, so is, like, everyone covering AI? How do you, how do you sort of split the work and, and all that? **Steph Palazzolo** [4:19] Yeah. No, it's funny. I feel like people always ask me, like, "Oh my God, AI is so, so crazy. How do you cover it by yourself?" And I'm like, "We actually have, like, 10 different, you know, 10 different writers who are basically covering, like, different areas of AI." So I, I get a lot of help. But yeah, specifically, so The Information, we've been around for over a decade. Um, we are kind of like a tech and finance media publication. Uh, specifically I would think of us as, like, yeah, kind of going above and beyond to get you behind the scenes and to get the scoops around what's happening at some of the most important, you know, tech media finance companies today. And so specifically, like, for me covering AI, I'm kind of part of, like, what you might think of as, like, the enterprise tech team. And so we have reporters who are covering, like, the major tech companies like Google, Amazon, Microsoft, but then we also have reporters that are covering, like, cloud more broadly or chips or, like, crypto. And so we have, like, a mix of reporters that cover specific companies, but then also reporters that are covering, like, larger beats, like AI for instance. Um, and then of course we have a- another part of the business that covers, like, finance and deals and retail companies and consumer companies. And so yeah, basically, like, we just cover anything and everything around tech and, and finance. So it's quite a wide array of things. And we're always, like, looping in, like, you know, anytime I do a story around, like, OpenAI's need for compute. Like, we're always looping in, like, my colleagues Anisa, who covers cloud, or Chenner who, who covers chips, and our reporters who cover, like, the big cloud providers. So it's very collaborative. **Swyx** [5:51] Yeah, I'm sure it has to be. Uh, okay. You, you, you, so you... Let, let's, let's see what you do, right? You, you, you write the AI Agenda, you show on, uh, you, you, um, also appear on TiTV, which is, like, your new sort of TV channel. You, you report to groups like, like once to twice a week I, I feel like, uh. How do you... Like, is it... You know, how do you do it? What, what is- what else is, um, part of your role? **Steph Palazzolo** [6:18] Yeah. Yeah, no, it's kind of funny because, like, I think for me, 'cause I came from finance, I never had done a journalism job before that. Like, I still feel like I'm pretty new to the field and learning about how things work, and whenever I first joined, I think you kind of just assume that as a writer, like, most of your time would be spent writing. But that's actually probably, like, 20% of my day. Um, I think, like, the vast majority of my day is spent maybe like, I don't know. I don't know if VCs would like this analogy, but maybe in some ways, like how a VC might, where you're constantly reaching out to founders and interesting startups and asking people like, "Oh, if I want to learn about X-Y-Z topic, who should I be talking to next?" So, like, the vast majority of my day is actually on, on calls with, with, with, you know, founders and researchers and investors, and basically just, like, begging them and being like, "Hey, like, what's, like the hottest gossip that you guys know?" Or like, "What's, like, the coolest new trend in AI that, that you've seen?" And so I think it's just constantly trying to stay ahead of what's in the mainstream news about AI, and I think for me, two things that I really care about are, you know, maybe you relate to this, like previously working in finance is thinking about, like, the business models and the financials of these companies. So we've done a bunch of pieces or, like, I've done a bunch of pieces around, like, inference pr- providers and, like, their gross margins and business models, which, which initially might sound very boring but I think is actually really interesting. And then also, like, on the tech side of things, like, I think we're always trying to understand what is the trajectory of AI development. Like, what are the obstacles and how are the big AI labs trying to get, you know, past those, those challenges? **Swyx** [7:55] Yeah, we can get right into it. I, I was actually gonna start with the labs first, but, I mean, since, since inference is coming up, uh, today's big news is Modal is raising a billion dollars. I saw the other day Fireworks is raising at $4 billion. ### Inference Boom **Swyx** [8:06] Things seem on fire everywhere, but like, you know, what's going on in the inference market? **Steph Palazzolo** [8:11] Yeah. **Swyx** [8:12] Yeah. **Steph Palazzolo** [8:13] Yeah. **Swyx** [8:14] What's the, what's the high level take, uh, for, for people- ... people kinda out of, out of the loop? **Steph Palazzolo** [8:18] Yes. The high level take, so yes, basically I'm sure a lot of your listeners already know this, but these inference providers are companies that are, you know, helping developers, like, run and train and customize open source AI models more easily than they would be able to otherwise. And so yeah, there's just so many companies in the space. You mentioned some of them, Fireworks, Modal. There's Together, there's Base10. There's just a ton of these companies in the space- **Swyx** [8:40] Fal.al **Steph Palazzolo** [8:41] ... and so- **Swyx** [8:42] Fal for- **Steph Palazzolo** [8:42] Yeah, Fal's another one **Swyx** [8:43] ... Fireworks. **Steph Palazzolo** [8:43] Yes. Um, and so yeah, it's, it's very interesting because a lot of these companies have gotten so much funding, but there's still a lot of, like, skepticism around this space and, like, whether these companies are actually gonna end up being like, you know, like billions or trillions of dollars... Like trillion dollar company. Um, so I think a lot of people, like, think... Basically just call these companies, like, GPU resellers, so they're just like, "Oh, all you're doing is just being like a cloud basically, you know, helping developers get access to chips to run models on. So you're like a knockoff Amazon or like a knockoff, like, Azure or Google Cloud," um, which is, like, a bit derogatory. But I think that's kind of, like, what the con argument... That's, like, the argument against them, um, because also unlike software companies, they have to, like, spend so much money getting the, you know, getting those chips, and they have to keep a bunch of them, like, unused at any given time in case there's a sudden spike in demand. I think, like, the argument for them that I've heard from investors is just like, yeah, like, as, you know, AI apps become more popular, like, we're gonna see the amount of money going into AI inference to just, like, explode and go way beyond the amount of money and compute being spent on, on training. So we should have a bunch of companies that are available here to, like, help developers build these apps and, like, access models really quickly. And I think the space is also... There are some companies in the space that are starting to, like, think about maybe having their own data centers and moving more into, like, the data center, like, truly cloud provider world versus just, like, I'm gonna use chips from existing clouds. So yeah, I don't know. It's, like, a very interesting space, and I think it's just an example of, like, a new category of startups that really didn't exist, like, even two or three years ago that are just popping up because of, like, the demand for AI models. **Swyx** [10:34] Yeah. I mean, and, you know, I, I think just to point to another example that we didn't mention, but CoreWeave, technically a GPU reseller, um, and you know, a public company. So, like definitely a lot of room. **Steph Palazzolo** [10:43] I mean, I, I mean, I, I would be curious for Alessio, like, what you think about this space, like, from kind of like the, the VC side of things. Give us your hottest take. **Alessio** [10:53] I think on the VC side initially, I think everybody-- Well, not everybody obviously because some people invested in it. I think the first... The timeline of when the existing companies were gonna catch up, I think people overestimate it, how quickly they will do it. I think the same thing happened in coding, which I think the, a lot of the sentiment initially was like, well, you know... I mean, we had Cursor in our pockets like two and a half years ago. **Steph Palazzolo** [11:13] Mm-hmm. **Alessio** [11:14] And the sentiment is like, oh, it's great. OpenAI is using it, but, like, GitHub cannot possibly let them run away with it, you know? And I think with inference it was, like, similar, where it's like, hey, you know, obviously these tools are growing quickly because, you know, the main providers don't really offer these and um, blah, blah, blah. And then there was kinda like this question of is there like a AI air pocket in pre-training where you kinda have all these, like, commitments on the GPU and the hyperscalers, but then they're gonna go back to inference and so it's kinda gonna bring the price so down. Can you really build a company in the space? But I think Fal especially has done amazing because, um, media is still very open source driven when it comes to image and video generation. In a way the language is not as much. So I think they- they've been doing great and, you know, very happy for their announcement today. But yeah, we haven't done any, but also, like, we have a smaller fund and I think all of these inference things are, like, big money. **Swyx** [12:06] ... investments. Like you cannot do a $5 million round in like an inference company, you know? You kinda need a good chunk of capital to, to run the business. So obviously, uh, luckily the VC model has evolved now to have these mega funds that are able to support these like 100 million... Like, you know, Fal raised 125 million, million today. You need a lot of money to run these things, and to get a raise- **Steph Palazzolo** [12:26] Yeah **Swyx** [12:26] ... a huge amount of money, like it's not easy to survive. **Steph Palazzolo** [12:29] Yeah. **Swyx** [12:29] And I, I'm curious to see where, where they'll land in a, in a few years. **Steph Palazzolo** [12:32] Yeah. No, it's crazy to just see like casually like $100 million plus round of being like, "Oh yeah, that's so... that's totally normal for an early-stage AI startup." So yeah. **Swyx** [12:42] I mean, 2 billion seed rounds is normal now, so you know. **Steph Palazzolo** [12:45] That's true. That's true, yeah. The, the small, the small seed round of, of $2 billion. **Swyx** [12:50] I mean, we- we'll, we'll get, we'll get to the labs, but I just wanna stay focused on, on the inference side. Uh, we did have a chat with Eric, uh, from Moldo, and you know, we talked a little bit about how it's almost a little bit like a bank, right? Like you... It's basically what you're doing is maturity transformation of, of long-term contracts into shorter term, uh, you know, usage of p- of things. And you know, I think who- whoever is the best technologist, uh, and the best developer experience is gonna win there in terms of serving out these GPUs for providers. And the other thing I always think about from a finance point of view that maybe people don't even think, don't think about that closely, but obviously push back if you, if you disagree, is that it's basically a proxy for the gap between open models and closed models. Because if the open models do better, um, sort of the API hosts, the inference hosts do better. And if open, open models fall behind, then they do worse. So, so far, DeepSeek obviously like still one of the biggest news of the, the, the year. What was a big gift to, to, uh, to, to the, the inference providers and the sort of relative decline of Llama and disappointment Llama 4 was, uh, you know, a, a, a, a, a problem for them. So yeah, that, that'll be how I characterize it. **Steph Palazzolo** [14:00] Yeah. Yeah. No, I think, I think that's a very fair point. I do think there's probably a lot of usage of the inference providers for not necessarily to get state-of-the-art models, but because like there's so many tasks that you can... that you'd wanna use like a cheaper model for. Or even... I know even whenever like the DeepSeeker R1 model first came out, one of the major like startups in the space was like, "Yeah, 75% of the usage that we see for, for R1 is people trying to distill it into like-" "... whatever specific use cas- uh, u- use case they want," which I think is like really interesting 'cause I don't know if I would think that like distillation is like a... Like I, I don't know if I would say it's a good thing that 75% of your usage is distillation, but I mean, you can't deny that like whenever you see models come out like DeepSeek, like that, that is a huge draw to these inference startups for sure. **Swyx** [14:48] Yeah. Um, I, I think that's absolutely it. Um, um, and also partially because DeepSeek was huge, right? Uh, there's a lot of these like Chemi, K2, and what's the other one? GLM, uh, 4.5 right now. All of these are hundreds of billions of parameters, like very costly to, to serve for like sort of day-to-day use, so you have to distill it somehow. **Steph Palazzolo** [15:06] Yeah. That's fair. **Swyx** [15:07] Cool. Yeah. Uh, so the, I think the other more recent news... I'm just gonna go in sort of like reverse chron order, but also like feel free to just do tangents on, on themes and stuff. Uh, the next one is Meta. ### Meta's Bet **Swyx** [15:18] This is a fun one. I was super happy to reach out to you because you not on- only like breaking news and scooping things, you actually put opinions into your newsletter, which is I think very brave as a journalist, right? **Steph Palazzolo** [15:30] Thank you. **Swyx** [15:30] Because like sometimes you have to like meet these people face to face, and you're out here saying Meta should not have super intelligence. What's up with that? **Steph Palazzolo** [15:38] That's true. Yeah. No, whenever I... whenever we published that with that headline, I was like, "God, I feel like I'm gonna get some angry comments with this." No, but like... I don't know. I think there's like two ways to look at this. Like I think one way is like obviously if we're going to reach super intelligence, like I think every big tech company with the resources like does not wanna miss out on that. So like it makes sense. Like Meta's like, you know, if super intelligence is the future, like I wanna be part of that and I want to be part of like that development. But I think just if you take a look at like how Meta actually makes money today, right? It's through like social media apps and ads and these things that are... have been around for, for quite a while. Oh, yes, there's my... I love looking at these. **Swyx** [16:25] I'm providing visual aid for people- **Steph Palazzolo** [16:27] Thank you **Swyx** [16:27] ... and also people can Google it. Yeah. **Steph Palazzolo** [16:29] Thank you. Yes. But like if you think about it, a lot of things like ads, like you don't need super intelligence. So whenever you think about all the investment that Meta's putting into AI today, which is great and great for a lot of open source developers, like maybe what's actually good for their P&L and their finances is to not focus so much on building these like insane, huge state-of-the-art models, which they've obviously struggled with more recently. But it's to take things that are, you know, 70, 80, like 90% as good as the best stuff out there and figure out how to use it in its apps. Like I think that's like a problem I've seen from like Meta and a lot of other companies where it's like the models are getting better, that's great, but it's like if I'm going on Instagram and like the only AI stuff I see is like, oh, do I wanna talk to like, like an AI version of like the Hawk Tuah girl or like... Like I'm like, "Who's using this?" Like like this cannot be like where consumer AI ends up, right? Like that's just so depressing and like I don't know who's using this stuff. So I'm like I feel like we've seen progress in models and even for Meta, like maybe the... I mean, as we saw with Llama 4, even the progress wasn't like amazing it seems, but like we've seen even less progress like in the actual products and like innovation there. So I'm like, I'd love it if Meta's, you know, came up with some really cool like AI apps that I couldn't even think of versus just like more and more models. But yeah, I'm curious like what you guys think, and I think people have very strong opinions on whether Mark is making the right moves with like some of these big hires and the super intelligence labs and all this craziness going on. **Swyx** [18:09] Uh, uh, yeah, I mean, it's really hard to know what to think. Uh, obviously, so we're, we're recording this the day after Facebook had their earnings report, and Mark, uh, put this, like, personal letter, which was very inspired by Nat Friedman- ... on personalized super intelligence. I host the AI Engineer Summit in World's Fair, and at the summit in, in, in February, uh, Sumir from Meta was also talking about personal intelligence, which I think, like, it, it, it's, it's of a kind. Like, I think, um, that part of that group in that, that part of Facebook really believes in that, that agenda. We're hoping to talk to Mark, uh, for the podcast. We'll, we'll, we'll ask him directly. Uh, a lot of people think that it's basically he just wants to be the dominant player in every platform. So like, you know, if VR is a thing, fine, we'll do VR. If crypto's a thing, w- fine, we'll, we'll have a crypto play, you know? And, like, this is, this is the, the, the AI play, but obviously I think he's going after this bigger than basically everything else he's ever done, which is his right as a, as a young founder still, still in control of his company. But, like, it's, it's also very bold and it's very inspiring. I mean, we haven't shipped anything yet, so we don't know what, like, what, what it's gonna look like. But I think it makes sense. I don't know about business model, but, like, you know, he was happy to bleed, like, $20 billion a year on, on VR, so, like, why not? **Steph Palazzolo** [19:15] Yeah. **Swyx** [19:15] They got glasses. They just put AI in the glasses and it's fine. We're all gonna wear glasses. **Steph Palazzolo** [19:22] Yeah. **Swyx** [19:22] Yeah. **Steph Palazzolo** [19:23] Yeah. I mean, it was... I mean, even, like, reading his, the kind of letter he put out about personal super intelligence and, like, hearing the way he talks about super intelligence is very interesting in contrast with, like, you know, like, Anthropic, OpenAI, these other players that are like, "We wanna solve, like, the world's most dangerous, like, diseases and, you know, solve all these, like, huge problems we have in science." And then, yeah, it, it's interesting. I think Meta has really just doubled down on this idea of like, "Okay, like, that's cool and all, but just, like, how about the everyday person? Like, we wanna help them too," which I think, I think is good. I think it's good that they're not trying to just do what the other labs are doing. But I feel like I still have to see how the personal super intelligence, like, marketing and angle is gonna play out with, like, consumers and stuff. **Swyx** [20:05] Yeah, totally. Well, the other big part of MSL, which obviously everyone's very excited about, is the pay packages that people are getting. You guys have been reporting very diligently on that. Another... This is also, like, very breaking news. I'm glad that we're doing this today rather than, like, maybe last week because I, I get to up to, like, update things, right? And we get, we get some fresh perspectives. Me- a Meta po- spokesperson said, uh, pushed back on, on the sort of billion-dollar offers that have been rumored, right? Uh, so, like, we're past the 100 million mark right now. We're at, we're at a billion. Um, and he, he-- it seems, seems like he deleted this tweet, but basically he, he sort of pushed back on the, uh, the, the sort of pay packages. But, like, how do you know when you're being told the truth, right? You have a lot of people leaking you information. There are a lot of people accusing Sam Altman of, say, of throwing out numbers in order to play the media. I don't know how true that is. It could be true, it could not be. I just, I just don't know. Like, how do you know when you're being played? **Steph Palazzolo** [21:02] Yeah. Yeah. I mean, it's a, it's a tough... Yeah. It's something that I think every journalist has to struggle with and deal with. I think first is thinking about the incentives of the person that you're talking to. So obviously if I'm talking to a PR person at OpenAI or Anthropic or one of the big labs, like, they're obviously going to be pushing a certain narrative that's, like, probably positive about their company and maybe not so positive about other companies. But even, like, some of the researchers and stuff, like, they might have very strong opinions about, like, the safety culture at An- Anthropic or, like, again, like, Mark's strategy at, at Meta. So, like, everyone, like, has biases and they wanna, like, make themselves and their companies, like, look good. So I think you just first have to keep that in mind. So I think, like, pretty much anything we get a tip like this where it's, like, a huge number that makes a certain company look good and another company not look so good, we wouldn't report that unless we had, like, multiple sources and we would know, like, who, you know, like, who did that offer go to? If it's over multiple years, how much of that is, like, cash? How much of that is stock? Is it over, like, two years? Is it over four years? Right? There's a lot of ways to, like, manipulate a number like 1 billion and make it seem, like, way more crazy or, like, way less crazy than what it actually is. So I think it's very tough because, like, there's crazy numbers like that getting thrown around. But, like, I think as journalists, we do have responsibility to get into the details and kind of question a little bit more, like, "Okay, why is this person telling me this?" And, like, is it actually helpful for me to write this number, or is there actually, like, X, Y, Z things that I also would need for this to be, like, a full, like a story that tells the full truth? So yeah, I think the reporting on some of these comp numbers has been a little tough because I think at, at this point, anyone's just throwing out any, like, hundreds of millions of dollars number or billion-dollar off- like, offer number and everyone's like, "Okay, like, that sounds reasonable." So, like, yeah. I think some of these numbers, like, I think journalists should include more details on, like, exactly how these offers are structured that could make them look better or worse. **Swyx** [23:07] Yeah, like I have Dylan Patel pushing back on even the $100 million si- uh, signing bonuses that people, that I think you guys have reported. Like, I, I don't know what the, the fact... I mean, unless you've seen the actual offer, like, it's really hard, like, you know, firsthand to, to actually know what's going on. **Steph Palazzolo** [23:22] Yeah. That's, that's true. And I, I do think, like, there are situations where, like, Sam Altman is on a podcast like, "Yeah," like, "Mark's given $100 million offers to our- **Swyx** [23:32] Right **Steph Palazzolo** [23:32] ... researchers." **Swyx** [23:33] Right. **Steph Palazzolo** [23:33] And right, and it's like that's immediately a headline because of, like, how important Sam Altman is. But, like, it doesn't really go through the same, like, scrutiny as just a normal news story from, like, a publication. So I think a lot of CEOs are very smart about this and they, like, know how to throw out stuff and immediately it catches on with- **Swyx** [23:50] Yeah **Steph Palazzolo** [23:50] ... the news. **Swyx** [23:50] Yeah, because, like, then they, then that opens for counter narratives. Which it's just, it's, to me, it's fascinating how CEOs sort of y- kind of use the media as a tool to play games with other CEOs. Obviously, and, like, we're part of it, like, so it's, this is a very- ... meta, meta thing. But, like, you know, like recently, I think this week also, there's, there's this story about, like, how, you know, multiple, multiple of these offers have gone to Thinking Machines people, and all of them have rejected it or whatever, right? And definitely that's a planted story from Thinking, Thinking Machines. I, I, I don't, I don't know how to, how else to say it. They just wanna look good. **Steph Palazzolo** [24:21] I mean, yeah, that story definitely makes Thinking Machines Lab, like, look, look good for sure. **Swyx** [24:27] Okay, cool. Before we move on from Meta, I just wanna, you know, leave the door open. Any other themes that you're watching, uh, on, on the Meta side? **Steph Palazzolo** [24:37] Yeah, I mean, I think we definitely covered a lot of the really interesting ones. I think-- I mean, one theme that I think like Mark has even kind of touched on recently is whether they'll continue open sourcing models, whether they're gonna start closed sourcing stuff, and like what does that mean for its business model? Um, 'cause obviously like, like right now, you know, Meta doesn't make money from an, an, an API the way that like OpenAI and Anthropic does. But if they start closed sourcing models, maybe that might change or that could change kind of the shape of the way that they think about how to make money from this. And so yeah, I think that's very interesting and, I don't know, I think for me as a consumer, like I'm always really interested in like consumer AI because I feel like, uh, there's not really as much coverage of on it as I would like and like I personally... I use a lot of AI stuff, but nothing that's like truly kind of like transformed my like non-work-related life. So I'm very curious like what Meta will come out with, with consumer AI. **Swyx** [25:29] And, you know, if, if there's any lab that will do it, it's them and OpenAI, I guess. Yeah, I, you know, we, we did a podca-- We, we also had our first, I think our first VCs on the podcast, Justine and Olivia Moore, who cover consumer AI, and that just seems like a fun job. They scroll TikTok for a living. That's, that's what they do. **Steph Palazzolo** [25:45] Oh, wow. Okay. Well, yeah, I, I also scroll TikTok for a living, so you know. I, I, I wish I could get paid money for that, but... No, but it's a very tough job. Honestly, I think investing in consumer AI is like in some ways way trickier than investing in enterprise AI, so props to them. **Swyx** [26:01] Yeah, I, I've come around a little. So I think that is the Silicon Valley narrative, right? That like consumer AI is, uh, kind of a crapshoot. Um, you know, sure, every n- every now and then you get a Facebook, but most, most of the time it, it flops. But I think like I think the, the, the idea or the meta or like discussion or vibe in Silicon Valley has shifted a little bit, mostly basically influenced by ChatGPT, where if, you know, if you build a strong consumer business, you can get the scale and the diversity to build up like infrastructure and then data and, and all that. And I think like it seems very powerful, but ob- obviously also subject to a lot of fads anyway. I, I'm not, I'm not a consumer guy, you can tell. Cool. Which-- what other lab is sort of top of mind? ### GPT-5 **Swyx** [26:38] I mean, I, I, I guess, you know, uh, we're, we're recording this on with, uh, GPT-5 imminence. That's another big story. Uh, you were on TITV talking about it. How has been your coverage of just generally just, you know, stepping back, OpenAI in general? Like a very tricky, large, secretive company to, to cover. What's it like? **Steph Palazzolo** [26:57] Yeah. I mean, it's, it's crazy. Well, we are hiring specifically an OpenAI writer, so that can tell you like how much work goes into covering that, that company and the fact that we, you know- **Swyx** [27:08] Yeah, you're only doing OpenAI? **Steph Palazzolo** [27:10] Yeah. So we wanna have one person that's just like reporting on OpenAI, which is like crazy. I feel like, um... Yeah, I can't even think really of past examples of where you have, you'd have a journalist, you know, focused on like one, you know, early stage startup just by, by itself. But yes, it's a very complicated and somewhat se-secretive company and I think especially because like there are so many angles to come at it from, right? Like there's like consumer product, there's enterprise side, there's, you know, the work it's doing with the government now and the, and the regulation and laws and, and policy that it's involved with. And then obviously there's stuff like Stargate and Compute and like the crazy data center projects that they're working on. Yeah, I mean, like really like any angle under the sun, like you can write about it with, with OpenAI. I think specifically whenever I think about like GPT-5, which we recently wr- wrote a story on, um, I think like what's so interesting about that to me is that kind of like, you know, the past year we've seen a lot of progress, especially in reasoning models. But I think like under all of that has been the fact that like GPT-4 has been kind of the like leading GPT model for, for a very long time and everyone's kind of like, "Okay, when is something worthy of the GPT-5 name going to come out?" Like we saw 4.5, which was like low-key kind of a flop and like I think has actually been deprecated from the API now, which is crazy. **Swyx** [28:37] Yes. I'm very sad. **Steph Palazzolo** [28:39] Yes. But like, you know, as much as we see o1, o3 doing super well, like this has been the big question, like what is going on with GPT? And the story that we wrote, we kind of talked about how at least the people that we've talked to who've tested it so far have been pretty impressed. They seem to think that it's been, you know, im- there's been improvements in a number of domains and both like, like science and coding, but also areas like creative writing and general knowledge. But I think like the question that we have coming out of this is how, like what does the performance of GPT-5 say about the trajectory of AI? So if GPT-5 is like a flop for some reason, like what does that mean about, like what does that mean for reasoning models and like the good progress that we've seen there? Like is o4 also gonna be a flop? But then on the flip side, if GPT-5 is good, is that necessarily because pre-training is still going super well and we're, we've ab- we've been able to scale up pre-training really nicely and everything's going super smoothly? Or is it actually because like reinforcement learning is the area that we're getting the most improvements from now? And like if that's true, how much further can we go with like reinforcement learning and, and post-training? So I think like that's gonna be the most interesting part coming out of the GPT-5 launch is kind of not just thinking about like, "Oh, like what does this mean for OpenAI standing in the AI race?" But like literally what does this mean for just like AI progress for every AI lab? **Swyx** [30:16] Yeah. **Alessio** [30:17] How do you think about poking into the other labs? So, you know, GPT-5 is coming out soon. Are you in parallel trying to see, okay, is Anthropic gonna release something? Is Gemini gonna release something? Is like, is xAI... You know, Elon always wants to drop things at the same time. What's kinda like the, the routine? **Steph Palazzolo** [30:33] Yeah. Yeah. I mean, we're always keeping an eye out on, like, all the labs, and there definitely have been some points where I know, like, a year or two ago, I think OpenAI, like, released their, like, multimodal model, like, the day before Google released theirs. So there's always, like, some fun little, like, drama happening in the background. But I think for the most part, at least, like... And I hope I don't jinx myself. For the most part, I think right now, like, GPT-5 is the main model on the horizon for us. I think most of the other labs, like, have, like, somewhat recently released models, and so I would expect maybe for it to be, like, a couple months before they come out with, with something. But, you know, like, we see Anthropic, for instance, like, raising a, a, a huge round right now. And so I have no idea if that's at all affected by, like, GPT-5 coming out and them wanting to, like, really accelerate things because they're, like, worried, for instance. But yeah, I think depending on how this model goes, like, we'll definitely see more models released from some of these other labs or, like, them wanting to raise rounds and things like that to, like, get the next model here faster. **Swyx** [31:32] Yeah. Uh, I, I, I mean, I think, um, weird cycle of you ship an big model, get some noi- get some, um, tract- get some, like I said, traction and noise and h- excitement, and that helps you with the fundraise for the next thing. You know, happy to move on, uh, in terms of topics, but obviously OpenAI is, is very huge. Uh, but like, I do think that it's also very interesting when each lab CEO puts out a blog post because it often tends to, tends to lead the fundraise. So I think the most directly, I think, for example, Machines Loving, Loving Grace was directly before the $100 billion fundraise for Anthropic. And, um, you know, I think, I think, like, th- there's... I think there's some correlation there in terms of, like, when Dario sh- shows up somewhere and says like, "Oh, like, a third of the workforce is going away," like, that's also, like, a very obvious, like, "I am fundraising" sign. And I, I don't know if, like, people read the tea leaves like that or am I, you know, being like sort of conspiracy theory. **Steph Palazzolo** [32:23] Yeah. Yeah. I guess I think about, what was it? The Anthropic memo where they're talking about taking money from, like, the Middle East. Although I think that one's, like, pretty obvious. **Swyx** [32:31] That was a bizarre one. **Steph Palazzolo** [32:32] That one's like, okay, definitely something's going on. **Swyx** [32:34] That was... What, what was that? Like- **Steph Palazzolo** [32:36] Yeah. That was very- **Swyx** [32:37] ... they didn't have to say that. **Steph Palazzolo** [32:39] Yeah. That was, like, very interesting. I mean, I don't know. Obviously, like, I don't know Dario personally or anything like that, but I think in some ways you gotta respect somebody who's, like, that open with their employees for better or for worse. Like, potentially for worse because it leaks, and I think people were, like, not super happy with them. Um, yeah. I don't know. I think it's interesting with, like, AI, there's a lot of these conflicts where it's like, you know, a lot of the companies, for instance, are, like, taking... are removing from their terms of usage, like, we can't... you can't use AI for weapons or military. Um, and then sometimes, like, you kind of hear from them, and they're basically just like, "Well, obviously I would-- I don't want our AI to be used to, like, kill people. But if we take this off, then at least we can, like, be in the room with, like, the military or the government and be able to, like, give them advice and, like, shape what they're doing." And so yeah, there's a lot of, like, weird, like, moral conflicts I think with these companies and, like, who they take money from or, like, who they allow to use their models. That's very tough. But yeah. What, what did you think of the Dario memo? Were you, like, surprised? **Swyx** [33:47] I, you know- **Steph Palazzolo** [33:48] What's that? **Swyx** [33:48] ... you see these things, but, you know, everyone knows, like, when you put something like this out, it's gonna be all over the press and, like, why would you put anything potentially embarrassing? Like, you just, you just didn't have to feed your, your critics. You know, you can just say like, "Hey, we're, we're, we're looking at the Middle East." You don't have to say like, "We're compromising our principles or whatever to, to go into the Middle East." Anyway, yeah, I, I think more broadly, uh, you know, I, I don't have, I don't have a take on that. I, I think, um, everyone has to sort of find their way. Um, I think Google, you know, has had to deal with the "don't be evil" slogan that they had in the early days and, you know, at, at some point, like, what does evil mean to you doesn't mean, doesn't, doesn't mean the same thing to me as if I live in a completely different country and context than you. So, like, yeah. Th- th- there's, there's one, there's one form of discussion. I think the other thing that, that happens a lot with big founders, with, with big CEOs, which, uh, uh, you know, I've had these sort of offli- offline conversations. A lot of them, like, want to build AGI, but, like, that, that I approve of in order to fight the AGI that I don't approve of. So, like, it's, you know, the only thing that can beat a bad AGI is a good AGI. And, and so they, they'd rather at least be part of that debate rather than, I guess, not be. So I think that's, that's a very motivational factor. **Steph Palazzolo** [34:53] Yes. Yeah. That's, that's very true. I feel like it's a very tech thing to be like, to be like, "I, I know how to protect us from, like, the evil tech." **Swyx** [35:01] I mean, you have to... Like, somebody has, has, has to try, and if you don't try, someone else will try. Um- **Steph Palazzolo** [35:06] That's true. Yeah **Swyx** [35:07] ... no matter what. So, like, you know, you wanna be high agency or you're not high agency. And, and the, the, I mean, the other thing is, like, you know, you, you ha- if you have the ability, then maybe you have the responsibility, you know, not to, not to be too Spider-Man about it. **Steph Palazzolo** [35:18] That's true. That is true. **Swyx** [35:20] So, so yeah, I mean, and then, you know, I think the other, the other things about Anthropic are, are like, um, you know, uh, they, they made a real dent with, with on the coding side. Are they going to be a coding-oriented lab? ### Coding War **Swyx** [35:34] They, you know, they, they have-- they made some movements on financial services and, and, uh, you know, other, other verticals like m- maybe defense. But I think coding is obviously a very key battlegrounds, like we just talked about how GPT-5 is maybe OpenAI sort of pushing back on Claude code and Sonnet and Opus. But, like, is, is that a big part of the conversations that you're having? **Steph Palazzolo** [35:53] Yeah, definitely. I mean, it definitely feels like Anthropic has decided that coding is kind of their, like, battleground, and I mean, so far they've been doing really well. I think, like, one thing I've heard, and I, I'd be curious, like, I think... Like, I'm not a developer obviously, so I'd be curious if this is your opinion as well. But, like, I think one thing that developers have told me is that, like, one reason why they like Claude for coding so much is because it's not only good at kind of like competitive or academic programming tasks, but it's also, like, these more, like, soft skill type thing-- or not, not soft skills, but more, like, more practical everyday things that they're good at, like working with, you know, huge code bases that have lots of, like, old legacy code in them or just even being, like, persistent with solving a coding problem and realizing, like, "Hey, I've tried X thing like three times and it hasn't worked. So, like, let me try something different." So it kind of feels like as scientific as, like, code feels, like- It seems like behind the scenes at Anthropic, there's a lot of like... It's like a like more kind of like art than science. Like there's a lot of like shaping of the model that is happening to help with kind of what engineers deal with during their day-to-day versus more like academic, you know, competitive programming questions. **Alessio** [36:59] I think the $200 a month for unlimited usage definitely helps people go towards Anthropic with Cloud Max on the Claude code. But yeah, the, the models are definitely very good at that. And I think they have now a great harness to get data to make the models even better. Like in the, um, in Claude Code now, it basically tells you, you know, you can respond and tell Claude what it got wrong if you don't wanna do the thing that it says it's gonna do. Um, and I think obviously OpenAI wants to get the same type of data with Codex, but I think they're way far behind as far as adoption, at least in our, in our circles. I'm sure they have a lot of usage in, in other groups. But yeah, the models are good. And, uh, you know, one thing Chroma released this context rod paper recently about context utilization, and the Claude models are actually the best at using kind of like longer context. So I think there's maybe been this question of like, you know, maybe the model is not better code necessarily. It's like just better using the context and, and kind of the day-to-day that ends up playing a lot, uh, a lot to it. So I think we don't yet know. Like you said, it's kind of like it, it's better, but like why is it better? Like it's unclear necessarily. **Steph Palazzolo** [38:05] Yeah. **Alessio** [38:05] But, um, the vibes are definitely, are definitely good. **Steph Palazzolo** [38:09] Yeah. It's like it's hard to explain in, in words. I mean, like how, how upset would you... I mean, I know Anthropic has come out with some new news about like kind of rate limits, but like how upset would you guys be if they like raised the pricing for from two hundred or like put some pretty intense like- **Alessio** [38:24] Mm-hmm **Steph Palazzolo** [38:24] ... usage-based pricing on it? **Alessio** [38:27] Well, you can pay as you go on top with the API, so at least, you know. The... I, I think that's what they're gonna test is like, okay, how much over do people go, uh, with the API as you go pricing? I don't know. I mean, I've always said with Cursor, I was always saying this, I was like, "I would definitely pay two hundred bucks a month for Cursor." Like when it was like twenty bucks a month, it just like it doesn't make sense that I'm like, "Oh, it's twenty." I don't know what the Claude number would be. I think, I think, yeah, four hundred I would probably pay. **Steph Palazzolo** [38:55] Okay. **Alessio** [38:55] But above that it's kind of like, okay, I don't know. But maybe once I start using the API as pay as you go on top of it and I realize how much I spend, I'm like, "Okay, maybe I should still doing it." I don't know. It's gonna be, uh, interesting. I, I don't know, Shawn, what, what your take is. **Steph Palazzolo** [39:11] Yeah. How much are you willing to pay for, uh- **Alessio** [39:13] Yeah **Steph Palazzolo** [39:14] ... Cursor or, you know. **Swyx** [39:15] Look, like, I mean, I'll pay as much as I, I get value out of it and if, you know, if I... my job is to produce a lot of code then and there's a lot of, you know, there's, there's some, there's some top tier thing that is generally not just producing slop but like actually productively paying, creating, creating code, then, then I think, you know, I would, I'd definitely pay like thousands a month, right? Like, uh, I was out when they came out with the sort of two hundred dollar a month plan for ChatGPT Pro. I for-forget the actual title, the names of these plans. I was, I was out here saying like the, they're, we're gonna have two thousand on ChatGPT, right? They're... Like they just have, they, they like it's just like stuff enough product in there and model value in there, you'll get it. And I mean, similarly I think like Devin is, is, uh, or sort of combinations Devin is like in the five hundred to a thousand a month, uh, tier and like I, I think prices will go up as long as you're using that inference productively. I think specifically for Anthropic and Claude Code, uh, it's really more about quote unquote the margins which is the, the meme I always post now every time there's a sort of Anthropic post about like how, you know, their revenue projections have gone like from four billion this year to nine billion, uh, because of Claude Code. **Steph Palazzolo** [40:18] Right. **Swyx** [40:18] And, um, and I always post like, "What are the margins?" Like and you guys started talking about top line, but like, you know, if you're, if you're all negative margins, then like, you know, it's not that exciting. **Steph Palazzolo** [40:25] Yeah. **Swyx** [40:26] But like- **Steph Palazzolo** [40:26] Yeah, no, I mean, we, we had a story yesterday about OpenAI and how like I think they've reached about twelve billion ARR and yeah, but their burn went from like they projected like one billion to like eight billion or something. So yeah, so like the question is like your revenue's great, but like how about those margins? So that too- **Swyx** [40:43] Yeah **Steph Palazzolo** [40:43] ... it's very fair. Yeah. **Swyx** [40:45] Yeah, a-and like it, it's gonna be the major factors of like compute personnel is now an increasingly key one, um, data. You know, we, we have this concept of the four wars of, of AI and it's, it's all around like the, the key battlegrounds that, that people have. Uh, the last one is multimodality. And so like, uh, I, I think like there, there's, there's a lot of like really interesting, uh, work there that's, that's being done that like, uh, y- I think it's like a balancing act. Like you, you just kind of have... You need a visionary founder of successes every like, you know, two to three times a year, um, and you just kind of keep the, the plate spinning while, uh, while, while you're growing this thing. But like I really wonder what the sort of quote unquote terminal value, you know, just to use the finance term of what all these things are. **Steph Palazzolo** [41:27] Yeah. Yeah, no, I mean, it's especially now that like, I mean, we, we saw the like boomerang with the Claude Code leads like going to Cursor and then coming back. **Swyx** [41:36] Okay, yeah. What was that? What was that? Can you double click on it? **Steph Palazzolo** [41:38] I-- To be honest, like I, I know that we like we kind of broke that story, but I think your guess is as good as mine. We know that they, that they went there, they were there for I think about two weeks and they came back. I think our sense is that it wasn't a situation where it was just about comp where like Anthropic was just like, "Oh, we'll like pay you more." But yeah, I don't know. It's a really interesting question. I don't know. **Swyx** [41:59] Okay. So for example- **Steph Palazzolo** [42:00] Yeah **Swyx** [42:00] ... you're saying it's not about comp in the sense that like they were poached by Cursor and then Dario raised this hundred billion dollars and he goes, turns around and goes like, "Hey, where are my Claude Code guys?" And they're like, "Screw you. Give them back." **Steph Palazzolo** [42:10] Yeah. I mean, I think, yeah, my, my sense is that it wasn't that case. I mean, I'm sure Cursor gave them like a great package and I'm sure like the package... Yeah, I'm sure we, we heard some of those numbers we'd be like, "Wow, that's, that's awesome." But I mean, like Anthropic like has a lot of money and like they're, they care a lot about their workers. So I think it was just about comp, like they could have probably just given them a higher offer if they, if they really wanted to keep them. So I think there is s-something else going on. I don't know if it's like culture or just like how interesting the work is. Something that must have brought them back to Anthropic, I think. **Swyx** [42:42] Margins. Like this' the other thing, right? Like it's, it's very damaging to like for somebody to go in, look at Cursor and go, "Yeah, I'm not sticking around in any eve," right? So yeah, I, I, I think to, to be super clear to anyone listening, I... this is not insider information. This is just, uh, you know, this is, these are the things that people say in the Valley. Of like how it looks when nobody talks about the actual reason and people just speculate. **Steph Palazzolo** [43:04] Yeah. **Swyx** [43:04] You did have an, a nice piece on like the retention curves of, uh, of, of these different coding agents up there. That's a big story. **Steph Palazzolo** [43:15] Yeah. **Swyx** [43:15] We can talk about that. We've, we've kind of segued it from, uh, the, the, the model labs into the, the coding agent startups. ### Windsurf **Swyx** [43:23] What was your sort of summary of like the whole Windsurf situation? Let, let's just leave it at that, the summary of Windsurf. **Steph Palazzolo** [43:31] Yeah. Summary of the Windsurf. Yeah, that act- actually is really funny 'cause that happened like while I was on a, on a boat in the Mediterranean, so I was like... I just came back and I was like, "Oh, my God," like "some, some shit went down while I was gone." No, but I mean, it, it's very interesting. I think it's another example of these like strange acqui-hire deals that are happening. But like I think it's interesting, like Windsurf, I think got so much more attention and honestly like hate from people in Silicon Valley, people on Twitter, on X, right, than some of these past deals because of like the, the structure and I think how many people were like left behind. And so... And I think there's also the sense of like a lot of these acqui-hires have happened when the startup that's getting acquired is maybe not doing super well or like it's kind of reached its peak and it's, it's on the way down. But it kind of felt like Windsurf was like on the up and up, and people were still really excited about them, and maybe they just got a deal that was like too good to turn down. But I think a lot of people were disappointed that like a company that was doing so well would kind of just choose to go like the acquisition route. Um, but yeah, I mean, I don't know, like as developers, like, I mean, I'm assuming that you guys would prefer Wind- maybe you'd prefer Windsurf to stay an, an independent company so you have more like coding assistance and options to choose from. Maybe that doesn't matter, but, um, yeah. **Swyx** [44:51] I think Windsurf as a, as a product is gonna stick around and like people who really like Windsurf, uh, I mean, I was a use- Windsurf user for a long time, um, uh, we are gonna keep using it because it fills a n- a need and like obviously Cognition bought it for, for a reason. Yeah, so, uh, no, no, no particular thoughts there. I think the, the people are mostly are in drama about, one, the sort of weird, uh, I've been calling this execu-hires, basically. Like, it's like not quite an acu- acqui-hire, but like you hire the executives. And then two, like, uh, obviously the, this quote-unquote, "the startup founding engineer employer contract," uh, which, uh, was, you know, uh, is, is calling the question now with, uh- **Steph Palazzolo** [45:27] Yeah **Swyx** [45:27] ... what happened. **Steph Palazzolo** [45:29] Yeah. **Swyx** [45:29] I mean, I think the, the, the thing that is uncertain for me is like, and there's been like five of these, Character, Scale, all these. You know, is this the new normal? Was the Windsurf deal any different? Because somehow the others had a lot less outrage than this one, even though a lot of the, the sort of surface level details were the same. Um, y- and like I mean, I, I would re- really love to like talk to one of the lawyers involved in like these things. Like how, how did they, how did they negotiate the deal points? And like technically the entire company is cashed out, so like how do the investors feel about it? **Steph Palazzolo** [46:00] Yeah. **Swyx** [46:00] It's all, it's all super interesting. **Steph Palazzolo** [46:02] Yeah. No, I mean, yeah, I think, I think two things. I think first like talking to investors, I think they're starting to... I think they're kind of, they're kind of over these types of deals. Like they're like, "Hey, I mean, like this is like fine, but like we don't invest in companies so they can get like a weird acqui-hire situation a couple years later. Like we're investing in them 'cause we think they can go public or like, you know, at the very least get bought and like a pretty good exit." So I think a lot of VCs are like not super thrilled with these deals continuing to be like the new norm. But yeah, I think to the something that you said earlier, like there's a sense that whenever you join a startup, like you're gonna be compensated for the risk that you're taking on of joining like something that's unproven, and then now you're having these situations where like, yeah, some people do get rewarded for taking on that risk, but then other people who have been there for like just as long or, you know, like feel like they've put in a lot of work, like are getting like seemingly nothing out of the deal. So yeah, I, I think it's definitely like messing with the trust between startup employees and CEOs and founders for sure. **Swyx** [47:03] Yeah. You know, I think the, the one thing I would sort of put a, a footnote in there is, um, you know, they like, they, they... I think even in the Windsurf deal, they, they did get, uh, offers at, um, at Google, but like, you know, there was some kind of revesting thing going on. Uh, yeah, cool. A- any other sort of, uh, uh, coding agents commentaries while we're still on the sort of coding topic? Anything else you're watching? How do you cover, uh, I, I guess like coding in general if, if you're, you know, out in, in New York and, you know, we're over here? **Steph Palazzolo** [47:35] Yeah. If I'm not, if I'm not like a developer and I'm like trying to understand what the heck is going on with coding startups. **Swyx** [47:41] Yeah. **Steph Palazzolo** [47:42] Yeah. I mean, I think a lot of it is just like finding sources and developers that I really trust and like being very straight up with them and being like, "Hey, I..." Like, "I'm not a developer, but like show me a couple examples of things that you built with this. Like tell me your opinions about what they're good for or not good for." I think like one thing I'm interested in is like the rise of some of these like open source coding assistants like Cline, which I, I know was on your podcast a while back. **Swyx** [48:09] Yeah. And they're announcing their, their aid today, yeah. **Steph Palazzolo** [48:12] Yeah. And um, and also the rise of kind of these like coding assistants targeting non-technical users like Bull or s- uh- **Swyx** [48:20] Lovable **Steph Palazzolo** [48:21] ... Lovable, where, yeah, you see a bunch of these startups that have raised a decent chunk of funding that are like, "We wanna target like either people wanting to make like prototypes of actual apps or just like small businesses that wanna make like a website or, you know, something that's not super com- complex, but like we'll get them up and running." Um, there's also been a lot of stories around like what can go wrong when you're using like vibe coding. Like it can accidentally delete your code. It can like totally not have some security thing that you need in there, and like lots of stuff gets leaked out. And so yeah, vibe coding is great, but I think we're only just now starting to see some of the consequences of what happens whenever you're not like careful with it. **Swyx** [48:57] Yeah, very mixed feelings about all those. Uh, go ahead, Alessio. **Alessio** [49:00] On the client thing, I was gonna say, sorry, it was just funny that Forbes did an article and used a laid in space bot screenshot as their **Steph Palazzolo** [49:09] Hi **Swyx** [49:10] Uh, for listeners, the screenshot guy is Alessio, so it's, it's actually kinda cool that, like, they stole his artwork, I guess. I don't know. **Steph Palazzolo** [49:18] Yeah, yeah. It was funny. Anyway, just, uh- **Swyx** [49:21] Uh- **Steph Palazzolo** [49:21] I feel like you guys should, you guys should get a, get a commission for that or something, you know? **Swyx** [49:25] It's Forbes, who cares? I, I mean, okay, I, I mean, I'll go there, you know, I don't know if it's offensive or not, but, like, look, like, Forbes, Forbes and Business- **Steph Palazzolo** [49:34] You've grown, you don't, uh, sorry **Swyx** [49:35] ... Fine, like, you don't work there anymore, but, like, Forbes and Business Insider, like, really fell off. Like, even, like, people are saying, like, TechCrunch really fell off. Like, The Information- ... is, like, actually up there. Like, we would never have those people on. ### Media Gap **Swyx** [49:44] We'll obviously talk to anybody at The Information. But, like- ... what's going on with, with, like, legacy online tech media has, like, really declined really hard and, like, I don't know what's go- you know, do you, is there- **Steph Palazzolo** [49:54] Yeah **Swyx** [49:54] ... is, is there chatter on the reporter side? **Steph Palazzolo** [49:57] Yeah, I mean, yeah. Media is, like, a very crazy industry. I mean, as somebody that, like, still feels pretty new to it, there's just a lot of stuff going on. I think, like, I think, like, business model-wise, like, ads are not working very, like, they're not working as well as they used to. I think, like, The Information, like, we do well because we have this subscription model, but, like, still I cannot tell you how many people complain to me 'cause they're like, "Why is your subscription, like, literally $500?" Like, that's insane. And so, like, yeah, so, like, we do it because we need to, like, stay alive as a business, and, like, I wanna get a salary. But, like, also there's, you're also giving up some, like, ac- accessibility for people who, like, wanna read our stuff. Although we do have many free newsletters, I'll put that out there. So there's, like, the business model stuff is, like, not going super well, and I think a lot of publications are struggling to kind of, like, find their niche in the way that they've done in the past. So, like, I think The Information, like, our whole thing is, like, scoops and exclusives, but, like, if you're a company that's more known for, like, summarizing, like, the daily news or even, like, having some sort of opinion or analysis, like, now that, you know, people on, like, X who are doing that or you can just use, like, ChatGPT to send you summary of the news every morning, like, you're just competing against so many other sources of news. And I don't know, I just feel like, in general, like, a lot of people just for some reason kind of distrust, like, media now and are turning to, like, influencers that they trust or, you know, people on social media, people on, on X for news because they feel like a personal connection with that person, and they, like, think that they are a trustworthy person. So I think the media's, like, fighting an uphill battle right now and it's, it's tough. And that's, like, tech media, so, like, I think, like, covering politics, like, way, way worse. **Swyx** [51:37] Yeah, yeah. Um, well, that's out of our domain, but I, I, I always think it's in- it's fascinating. Just, like, I hear, I hear some really awful stuff at TechCrunch and I'm like, well, not TechCrunch, you know? Like, TechCrunch is a, is an institution, you know? Anyway, so, so, okay, there's the, the two angles I really wanna focus in on. I'll, I'll start with the first one, which is, uh, scoops and exclusives, right? You guys are amazing at it. Could you peel back the curtain a little bit on how? You know, obviously, like, not the, the whole, like, secret sauce or anything, but, like- ... like, how, how, like, what, you know, what, how do you, how do you get people to talk to you? Like, I try to get there a little bit as an amateur, but, like, also you're really good at it. ### Scoop Craft **Swyx** [52:11] Like, let's- ... let's face it, like, you're really good at it. **Steph Palazzolo** [52:13] Thanks. I mean, I don't know about that, but I don't know. I think, like, I think everyone kinda has their own approach or, like, their own just, like, thing that they do. So I feel like I know some reporters who are, like, super blunt, like, very to the point, and they will just, like, be super persistent and just, like, call you up, ask you straight about the deal, and kinda give off the vibe of, like, "Hey," like, "I know what I'm talking about, and, like, I'm gonna get this information, so, like, you should help me." I think I know other people that are much more like- **Swyx** [52:40] That, this is, it's a caricature, right? **Steph Palazzolo** [52:44] Sorry? Oh, yes. **Swyx** [52:45] That, I mean, that- **Steph Palazzolo** [52:45] Exactly **Swyx** [52:45] ... that's the caricature model. **Steph Palazzolo** [52:47] Yes. And then also I know other people who are much more, like, kinda, like, uh, disarming or, like, they say stuff like, "Oh," like, you know, "I think I saw, like, this, like, tweet that said this, and, like, did you hear about that?" Like, I don't know. I think for me, like, I cannot be, like, scary or aggressive. Like, I unfortunately just don't have that gene, um, which I, I wish I, I did have it. So I feel like my thing is m- mostly just to be like, try to be, like, super curious and, like, tell, talk about my newsletter where I do cover things that are not scoops and be like, "Hey," like, "I, I love a good scoop as much as anyone else, but, like, I also care about the tech and the science, so, like, let's just, like, start there and talk about that." So I think, like, you need to kind of figure out, like, what works best for your style and, like, who you are as a person, but then also think about, oh, the person you're talking to. If I'm talking to a researcher, maybe I don't wanna just come out and be like, "Oh my God, is, like, so-and-so buying so-and-so?" But just start with more of like, "Hey, tell me about your research," and, like, "I'm just curious about it." Versus, like, if you're, like, talking to a VC, it might be more transactional where you're like, "Hey, I heard about these five deals. Have you heard about them?" Like, "Hopefully I'm, I'm helping you, so can you help me out too?" So it's just a lot of different approaches. I think also it's just, like, not being embarrassed about your job and just, like, like, literally, like, I will just cold call people I've never talked to and just be like, "Hello," like, "I don't, I know you don't know who I am, but, like, please tell me this piece of information." And, like, sometimes, like, you accidentally call their mom instead of them or you, like, get yelled at on the phone, and then you just have to be like, "Okay," like, "that's just the way it is," and, like, "It's just my job and I'll just move on." So there's a bit of kind of, like, just accepting the fact that this is your job and, like, not caring if sometimes you, people are like, "What is this annoying journalist doing calling me?" **Swyx** [54:33] I'm surprised you actually cold call. Like, I will never pick up a cold call. **Steph Palazzolo** [54:37] I know, yeah. No, uh, it's something that I feel like, uh, especially, like, millennials and Gen Z, like, we hate calling people, so- **Swyx** [54:44] Yeah **Steph Palazzolo** [54:44] ... that was, like, low-key something that I had to get over 'cause I was like, I used to hate calling people and I was like, "I'll just text them-" **Swyx** [54:49] Wow **Steph Palazzolo** [54:49] ... "and, like, maybe they'll, like, get back to me." But no, you're just- **Swyx** [54:52] Okay **Steph Palazzolo** [54:52] ... like, aggressively calling people all the time now. **Swyx** [54:54] I, I have a, I have a new respect for your, for your work. Uh- ... well, also, also, like, do you have to, like, how do you protect sources? So people trust you and, like, they can get fired for, if we're, like, telling you things they're not supposed to tell you. **Steph Palazzolo** [55:05] Yeah. **Swyx** [55:05] Do you just, like, reword things? You get a second source, or you say many people close to the matter? Like, what, what's, what are the techniques here? **Steph Palazzolo** [55:13] Yeah. I mean, I think for me, like, protecting my source is, like, my number one priority. Like, there are times where I hold off on a story or, you know, I let people... And maybe somebody tells me something and then they're like, "Actually, like I'm really worried. I don't want you to write about that anymore." I'm like, "It's okay. Like, this is not that big of a deal." Like, like, I've done stuff like, like that because, like, I care about my sources and, like, them obviously not getting in trouble or getting fired. But yeah, there's like, there's lots of stuff, stuff that you can do. Like, I think we always try to get, like, multiple sources on facts or obviously, like, I think we have a responsibility to our readers and, like, we want them to trust us, so we try to tell them as much as we can about, like, how our source knows this information, so if, like, they were in a, in a meeting or if they saw, like, a document. So we, we try not to use stuff like, oh, like a person familiar with XYZ, 'cause that's, like, quite vague. You don't really know, like, what that means. But we still do try to... We try to give some information but, like, keep it as broad as possible. So, like, instead of saying investor, we might say, like, a person who's talked to the executives or instead of saying, like, a person that worked on this team specifically, you might say, like, a person with knowledge of this product or, like, a person or, like, an, an employee at this company. So it's, like, all this stuff that you can do. I think it's just, it's just, like, building trust with your source over time and, like, you know, like, I want them to trust me 'cause I want this to be a long-term relationship, so, like, even if we just talk completely off the record the first, like, 10 times, like, that's totally fine. Like, I, I'm, I'm in it for the long run, so I just don't want them to ever feel, like, uncomfortable or like they're putting their, you know, their job at risk because of me. **Swyx** [56:45] Yeah, totally. I mean, we do the same for people that talk to me and, and Alessio. Yeah, so obviously anyone listening if, if you have stuff for, for Stephanie, um, I'm sure she's easy to reach. A lot of people you see to use Signal or Telegram. I don't, I mean, I don't think it's that encrypted or, you know. But, like, uh, yeah, that's, that's a, that's a tool of your trade, right? **Steph Palazzolo** [57:03] Yeah. **Swyx** [57:04] Um, I think the other, the other thing for, for readers I think is also, like, a cool, like, peek behind the curtain of, like, how these things get formed. Uh, I don't know if you want to say anything to readers on, like, sort of how to read the tea leaves on, on the stuff that, that you do, but, like, um, yeah, it's, it's fascinating. **Steph Palazzolo** [57:19] Yeah. That's, that's a good question. I like to think that we're, like, pretty straightforward. Like, I think we are, like, pretty, like, straightforward and to the point with our writing and- **Swyx** [57:30] You are **Steph Palazzolo** [57:30] ... we try to, like- **Swyx** [57:30] I, I really appreciate that **Steph Palazzolo** [57:31] ... break it down as much as possible. Oh, thank you. **Swyx** [57:33] Yeah. **Steph Palazzolo** [57:34] Yeah, sometimes it's like, I remember whenever I first started covering AI here, like, trying to think about how to explain, like, model weights in, like, a layman's per... I was like, "This is, like, impossible." But yeah, I think for readers, like, I think, you know, I think one thing that we and every media publication can always do better at is, like, taking a step back from the individual scoops and understanding, like, what this all means more broadly. And so we, we try to write about that in our, in our newsletters on, like, our TiTV. But a lot of the times I think it's, like, really easy to get caught up in, like, oh, this model dev-developer raised, like, this inference provider raised, and it's, like, very hard to kind of think about, okay, like, what does this all mean? Or, like, what does it mean that five inference providers raised in the last, like, three months or something? So I think whenever you're reading, just keep that in mind, and we will, like, link to related stories all the time. So, like, definitely let yourself go down that rabbit hole of, like, you know, understanding how this story isn't just a story or scoop on its own. It's connected to, like, this long narrative of, like, the progress of AI over the past couple years. And just try to think about, like, what does this mean in this, like, broader story that we're trying to tell here? And so obviously we wanna do that for you as much as we can, but I think that's also just a way to think about reading these scoops that will hopefully, like, let you understand, help you understand more what's going on in the broader context of the tech world than just, like, a single story. **Swyx** [58:59] Yeah. I, I, I definitely encourage people, um, the way to elevate the conversation instead of just repeating what you just saw on Twitter is the, to tie it into a broader trend to have an understanding of why now, understanding of, like, the economic forces at play, agendas of different players. It's a lot of what I do. It, it's, it's just trying to read into that. Um, honestly, like, um, I have a little, um, chip on my shoulder as I don't th- I don't think we do it that well, but I'm kind of curious, do you guys maintain a list of trends we are tracking? Databases of stuff, but, like, I almost feel like I can't keep up of how many kind of theses I'm, I'm tracking in my head because it all comes through conversations, and conversations, like, in a, in a single conversation you can have like 20, 20 different things that you touch on. **Steph Palazzolo** [59:42] Yeah, definitely. I mean, yeah, that's, like, the hardest thing I think sometimes. Like, I think there's some, like, just organizational things that you can do, like we have, like, weekly meetings about AI or, like, deals that we're chasing. And so I think some of that also rests on our editors because they're not kind of like in the weeds chasing down stories all the time. Like, they're better at, like, "Okay, here are, like, the five trends that we're thinking of writing stories about, and, like, I'm gonna assign this one to you, Stephanie, and, like, let's talk about this every week." So yeah. And I think there's, like... I mean, there's, like, s- this is obvious with stuff I do, like, I have, like, a notepad of, like, all ma- all the major trends that I'm chasing. And so at any point, I kind of, like, know what the top five, like, priorities are for myself. So I think it's just understanding, like, you can't remember all 30 trends and tips that are going on, but, like, just keep in mind the top five maybe that you're thinking about, and think about how tips and information that you hear can tie into those. **Swyx** [1:00:34] I was hoping you have some magical AI tool that I can... Yeah. **Steph Palazzolo** [1:00:37] Sorry. Yeah. It's, like, journalists, journalist AI. Well, if somebody wants to make an AI startup for that, that would be great, although I don't know if, like, journalist is a very lucrative market for companies to go after, but I would love a tool. **Swyx** [1:00:51] I, I, well, I mean, yeah, you know, like, meeting note-taking, like you say that you're in meetings all day, I mean, that's a huge market. It's, it's basically Granola and Notion and, you know, I think there's, there's a, there's a long tail of others. There's Grip Tape, which I think is journalist focused, which is kind of cool. **Steph Palazzolo** [1:01:05] Mm-hmm. **Swyx** [1:01:05] That they're, they're a bootstrap company. But, like, yeah, I mean, I, I think there's a lot, um, I think p- people want to get on top of their personal knowledge management. They want to build their second brain. I was a big part of that movement, uh, before AI stuff. But yeah. And then, uh, more broad- more broadly, theme- thematic-wise, you know, it's literally like, uh, again, like going back to like finance days, like that's what I did as a, as a hedge fund analyst. I would, I would track like portfolio themes and like here's, here's, here's all the, you know, the trends, the, uh, the news items that are coming in that change our bullishness or bearishness on, on a theme. And like, uh, then you sh- then you have to express that in your portfolio, uh, through the stocks that you make. But like my portfolio or your portfolio here is your coverage, right? Like, and, and like every guest we pick is a guest that, you know, takes a slot for a guest that we do not pick because- **Steph Palazzolo** [1:01:51] Right **Swyx** [1:01:51] ... we wanna emphasize this theme. For us, like we tend to emphasize engineers and coding, and so our lanes are relatively well-defined, but like yours are very wide. **Steph Palazzolo** [1:01:59] Yeah, definitely. ### Pitching Press **Swyx** [1:02:00] Okay, cool. Well, how do you choose to cover smaller startups, right? Like, a lot of people want to get on and be covered by The Information. Sometimes you cover, uh, you know, a, a lot of it's like the, the big labs and of course, like huge amounts of money. But like occasionally you cover like Vercel, and today you're covering Modal. Like, how do you, how do you choose the smaller t- startup coverage? When should they reach out to you? **Steph Palazzolo** [1:02:20] Yeah. No, it's funny about Modal that we... It's like the world has gotten so crazy now that like a billion-dollar startup, we're like, "Oh, it's like a baby." But, um- **Swyx** [1:02:28] Like, "Oh, that's cute. But, but New York Tech, aha." **Steph Palazzolo** [1:02:32] True. True. **Swyx** [1:02:33] Right? **Steph Palazzolo** [1:02:34] Yeah, that was the key because I love New York. **Swyx** [1:02:35] The New Yorkers are like, finally. Like- **Steph Palazzolo** [1:02:39] Um- **Swyx** [1:02:40] Not just Datadog and MongoDB. **Steph Palazzolo** [1:02:42] Exactly. Exactly. Uh, no, but I mean, like I... Yeah, like, I mean, just reach out to me and I... I'm, I'm-- Even if I don't cover... Okay, so actually let me take a step back. I think if you wanna get coverage, part of it is like understanding that, like it's a long game. You might not get coverage right now, but like I might not have the room in the newsletter to cover your funding round right now, but like I'm always looking to meet new startups and like hopefully we can find a way to do coverage or have some sort of story that's around your startup later. So I think don't just like come in and be like, "Hi, we've never met, but like I need you to cover my like $3 million seed round," or something. But I think also, like another way to pitch, to like talk to me and pitch to me is like, again, what I just mentioned about taking that step back. And so, like even if I don't cover smaller startups like standalone... Like for instance, this week, I did a story about kind of the next step in data labeling and data curation, which are like these startups building like RL environments for, for agents. And so I listed out like, I don't know, like four or five startups that are doing that, and these are some that like haven't even like really raised funding. They're like, they're like just pretty small. And one of the startups I talked to, like we had like a really great conversation, and I included a lot of information about like their company in the story because like they were just super helpful and like clearly very like well-educated on the space. And so I think if you're like a startup and you're like, "Hey, I think like my space is really interesting, and like people aren't really understanding like why this is important," like don't think about just pitching your startup to me, but like pitch me on like why your entire space is important and like I will write something. Like, it's more likely that I'll write something about a trend or like a new industry. **Swyx** [1:04:15] Yeah. Um, and by the way, uh, I actually have to draw for a, for an in-person meeting, but I, I, I figured this, this conversation was great, so I just wanna keep it going. Yeah, so that's great. I think, uh, I wanted to get that on the record because like I think people want to work with you, but like on... You know, a- and they just don't know how, and sometimes, like, you know, like I, uh, people wanna work with us, and the worst thing is they get a PR firm that like barely knows us, spells my name wrong, and like, you know, like gets in touch- **Steph Palazzolo** [1:04:42] Yeah **Swyx** [1:04:42] ... with me with like a, a story that is like completely not a story. And like they know it, I know it, like it's just a formality. They're just getting paid to send the email. Like, it's so routine, and I, I think, I think people wanna know how to break through the noise and come to you with the good stuff because a lot of people are working on good stuff. So yeah. **Steph Palazzolo** [1:04:57] Yeah, totally. Yeah, I mean, like, don't... Like, just send me like an email. Like, don't have... Yeah, don't have like a PR team send me like a really generic email and like just totally get your company wrong. Um, I think it's like b- better if it comes from you personally and like I'll, I'll appreciate it a lot more. **Swyx** [1:05:12] Yeah. I, I'll even say like for the early-stage startup founders listening, I, I think most of your PR teams are kind of pointless unless you really wanna get on like CNBC. Most, most of us are like easily reachable. You know, like, like w- like we, we screen tons of requests and it's fine. Like it's, it's, it's like you, you're probably much more, uh, much better at pitching your own company than, than the PR team. And we make a decision like really quickly, like it's not, it's not that complicated. Cool. Like, uh, yeah, so mo- more broadly, any other sort of coverage areas or themes that are, are sort of big hits that you're, um, that, you know, y- you want to talk about? Like, I don't get a sense of like what your biggest hit is or like your general sort of thematic overview, if that, if that makes sense. ### Next Wave **Swyx** [1:05:54] You know, there's like robots, there's xAI- **Steph Palazzolo** [1:05:57] Sure **Swyx** [1:05:57] ... there's Perplexity, there's all these other things. Uh, what, what stands out to you as something that you really wanna, uh, chat about? **Steph Palazzolo** [1:06:03] Yeah. I mean, I guess two things. Well, I think first, like the first thing is like, I mean, this is what I'm most interested in, but, um, I did kind of touch on it with like our conversation about GPT-5, but just like the trajectory of AI progress and kind of like, you know, first there was like pre-training scaling, and then now there's like reasoning models and reinforcement learning. Then, like what comes after that? Like, is it RL environments for agents? Is it something that's different than the transformer? Like, I think we're always curious just like what is coming next. And even, I like, I think that's why I'd love to talk to more people at the big AI labs because I feel like they're really on the forefront of that. **Swyx** [1:06:39] They can't tell you. **Steph Palazzolo** [1:06:40] I think- **Swyx** [1:06:40] That's like a national secret. **Steph Palazzolo** [1:06:42] Yeah, yeah. And it's like, yeah, I think a lot of the times it's, it's like I'm not even trying to get like a scoop. Like, I'm just like as a person living in the world- **Swyx** [1:06:49] You're personally curious **Steph Palazzolo** [1:06:51] ... of AI. **Swyx** [1:06:51] Yeah. **Steph Palazzolo** [1:06:51] It's like, what do I need to be like excited about? And like what do I need to be worried about? Like, I, I'm just like genuinely curious, not as a journalist, but again, like as a human being living in today's world. I think like... I mean, this is not a topic that I would like necessarily write about because like our, you know, our audience is like business-focused. But like, yeah, I think the societal impacts of AI is like fascinating. Like, I mean, I have a younger sibling who's going to college right now, and like You know, like has no idea, like what they should major in- **Swyx** [1:07:24] Is it worrying? **Steph Palazzolo** [1:07:24] ... or like what job they should have because, like everything's just so up in the air. And like, I have friends who are working in like investment banking who are being asked to test like financial modeling software that could one day, like replace their jobs. And that's without even getting into the whole like AI boyfriend/girlfriends, like people on Twitter going crazy because like the AI is supporting their like manic episodes. And just like, yeah, like there's just so much like we're shaping the AI, the A- the AI is shaping us and society. It's like, yeah, I don't know. It's like... And I think sometimes like you hear some of the CEOs say stuff like, like Mark from Meta made a comment where he was like, "Like on average today, people have like," I don't know what it was, "like four friends, but we really have demand for like 12." Which is just like such a crazy statement to say. **Swyx** [1:08:17] I didn't, I didn't- **Steph Palazzolo** [1:08:18] Yeah. **Swyx** [1:08:18] I didn't, I didn't read that. Okay. Interesting. **Steph Palazzolo** [1:08:20] Yeah. So it's kind of like, yeah, it, it's like this back and forth between technology and society and I'm just like very curious like what the average college student is gonna look like 10 years from now or like what my job is gonna be like 10 years from now. **Swyx** [1:08:33] Hmm. **Steph Palazzolo** [1:08:33] And how I'm gonna like, like how my friendships and, you know, work relationships and stuff are gonna, gonna work, so. **Swyx** [1:08:39] I have a lazy joke, but it's not really a joke. Like, uh, you know, everyone imagines that everyone else's job is going away but theirs, right? So Mark Andreessen's like, you know, like, "Everyone's job is at risk, but like, but VCs will be fine." DoorDash is like, "Podcasting will be fine." So like, uh, yeah, I mean, uh, uh, then that's a lazy throwaway joke. I think the more serious take is like, uh, the ta- I, I think, um, AI and just technology in general accelerates the difference between the haves and the have-nots. And so there's like relatively recent college grads that, um, that are now getting the $100 million offers from, from Meta and there's also like, I think something crazy like 20, 30% unemployment from fresh college grads for like people who don't meet that bar. And so like I think it's just gonna exaggerate inequalities. You know, that's why people talk about things like universal basic income. Some, some amount of like... Sometimes like human, human society is just not made to adjust this quickly, and things have been accelerating and like we just-- our systems just don't adjust that quickly. Like we have like annual schedules, but these things ship like two, three, you know, major models a year. And like, yeah, I, I, I do think like at some point, like, you know, the autonomy of knowledge work- automation of knowledge work, I think really starts to hit the GDP and the, and society and like, I think that's, uh, that's really scary. **Steph Palazzolo** [1:09:52] Yeah. Yeah. It's kind of like the whole joke of like instead of being like, "Oh, like I'm related to the Vanderbilt family," kids in the future are gonna be like, "Oh my God, my great-great-grandpa was like an, like a, a founding researcher at Meta Superintelligence Labs" or something. And like that's gonna be the new like status of being part of the elite. **Swyx** [1:10:11] This is sad. **Steph Palazzolo** [1:10:11] Which is like kind of scary. **Swyx** [1:10:13] Uh, I traffic a lot of memes, so this is like literally the daddy-- "What, what did we do to get so rich?" And they were like, you know, "Your dad took like, you know, the, the Meta, Meta offer." Uh- **Steph Palazzolo** [1:10:21] Exactly. **Swyx** [1:10:22] Well, yeah, I mean, like so I mean, I, I, uh, uh, and, and you know, my positive spin on this is that, you know, we, we used to s- freak out about kids wanting to grow up to be YouTube stars and TikTok stars. Now they wanna aspire to be ML researchers. Maybe that's not a bad thing. You know, if you wanna do more STEM, it's not a bad thing, you know? So maybe they'll, they'll be that. Uh, a-and, and, uh, you know... Well, I m- I mean, I think coming back to like the sort of more technical stuff because, you know, that's, that's the, the sort of meat and potatoes of what we do. I think most people that I talk to are not seriously pursuing alternative architectures. There are some, uh, notable exceptions, primarily together with the Mamba architecture, um, recursive with RWKV. There's like the XLSTM that was, uh, created by separate operator. And so like there are alternatives, but mostly people are happy scaling it up. And sometimes, um, this is what we call the lottery ticket hypothesis. It doesn't matter what the, the global best is, as long as like everyone's optimized for one architecture, it's fine. **Steph Palazzolo** [1:11:22] Okay. **Swyx** [1:11:22] It's kind of like English isn't the perfect language, but you know, it's good enough, like and everyone speaks it. **Steph Palazzolo** [1:11:27] Yeah. **Swyx** [1:11:28] Like, you know, it's got its quirks. **Steph Palazzolo** [1:11:30] Yeah. **Swyx** [1:11:30] But like we, like we did try to design a perfect lang-language. It's called Esperanto, and nobody speaks it. **Steph Palazzolo** [1:11:37] Yeah. **Swyx** [1:11:39] So like the- **Steph Palazzolo** [1:11:39] Yeah **Swyx** [1:11:39] ... Transformer is the lingua franca, I think like makes sense to me. And now like people are just like scaling up data and like all other, uh, and, and techniques. I'm curious if, um, you're the fifth person this week to talk about reinforcement learning environments. Is that just like a theme? Like apparently there are like 10 startups all doing this and like Applied Compute- **Steph Palazzolo** [1:11:58] Yeah **Swyx** [1:11:58] ... is like the, the, the, the, the hypey one. It's like three kids in a basement worth $500 million or something. **Steph Palazzolo** [1:12:04] Yeah. Yeah, no, I mean, I, I feel similar to you. This was like me last week. I was like, "Oh my god, why have like four people brought up this like idea to me?" **Swyx** [1:12:12] Okay. Okay. **Steph Palazzolo** [1:12:12] I like need to do like a, a newsletter on it. But yeah, I think it's like, like this idea where it's kind of like, why? Like let, let's just like let agents and models like roam free in this like escape room copy of whatever app and let it figure out for itself like how to like, you know, put a lead into Salesforce or like do these different things. And yeah, I think it's like this idea of like having less human involvement in reinforcement learning, um, and getting it to be more automated. But I mean, from what I've heard, it seems like still pretty early and like I think people are still trying to figure out like how to actually get it working super well. But- **Swyx** [1:12:47] Okay. So are you, are, are you, you saying when you say it's still pre-pretty early, like I expect this to work. I expect this to work immediately. Like is, is it not working? **Steph Palazzolo** [1:12:57] No, no. I, I, I think it is working, but like, like I think it's just very compute intensive to have like an AI agent like taking a bunch of like- **Swyx** [1:13:07] Actions **Steph Palazzolo** [1:13:07] ... actions in like this environment and like a lot of them like don't work out, some of them do. So I think it's like, it's like quite expensive. And at least the way somebody explained it to me is like if I wanna make like a copy of Salesforce to teach an agent how to like navigate it, like it's not easy to just create like a copy of Salesforce. There's like a lot of different like buttons and like different databases it connects to and like... Yeah, I think it's just complicated to make it a realistic environment for the model to play around in. But I agree. I think it's like really interesting and very promising and at least the one-- the startups that I've talked to, a decent chunk of them are already starting to work with like the AI, the big AI labs and like these are quite early startups. So it seems like they are very-- the AI labs are, are very excited about this approach. **Swyx** [1:13:50] Yeah. **Steph Palazzolo** [1:13:50] So. **Swyx** [1:13:51] Yeah. I, I would say I, I hear the same thing. Cool. Thank you so much for taking the time. Any other parting thoughts, questions, call to actions? **Steph Palazzolo** [1:13:59] No, I mean, this has been super fun. Yeah, I would just say for all your listeners, like I am always open to chatting, even completely off the record. And yeah, I don't know. I just like find AI to be like fascinating and I love talking to people about it, so please email or, or text me and I would love to chat. **Swyx** [1:14:15] Yeah. Um, thank you for all the great work you're doing. Um, yeah. Well, I'm sure we'll chat again. **Steph Palazzolo** [1:14:20] Sounds good. All right, thanks. --- This library is powered by PodHood (https://podhood.com), the podcast website platform.