PSA: @lex_feeds had some great takes on the @delphi_intel mini-feed this week: 1) How AI Conquered the US Economy 2) Simulate the World - DeepMind Researchers Unveil Genie3 3) Foundation Models Can't Really Learn Continuously - can they? Bookmark it:

what's the core question deAI faces? imo, it's perfectly captured by @PonderingDurian: "do the benefits of low-cost latent compute at the infra layer, open source compounding at the model layer, and creativity at the application layer ultimately outweigh the costs of orchestration which the big labs mitigate by bringing everything under the roof of a single company?" the answer depends on how you weigh immediate friction vs longterm leverage. big labs are betting that orchestration costs are so painful that people will pay premium prices to avoid them. they're probably right, in the short term. but once orchestration is solved (breakthroughs by @PluralisHQ, @NousResearch, & @PrimeIntellect are making me more confident this will happen), open ecosystems can unleash collective intelligence at a speed + scale that no centralized lab can match. this is where decentralized AI wins.

If you read one thing this week, would suggest the below AI report: "From Data Foundries to World Models" It interweaves data foundries, context engineering, RL environments, world models & more into an approachable, yet comprehensive essay on the current edge of AI.👇(0/12)

If you haven't had a chance to read my video model report, now might be a good time...

Remain thankful and learn to become unboreable, and you have basically already won.

"The Upcoming GPT-3 Moment for RL" Interesting post by @MechanizeWork on the shift towards massive-scale training across thousands of diverse environments. "Similarly, we suspect the GPT-3 moment for RL will be enabled largely by a paradigm we're calling replication training. This proposed paradigm involves tasking AIs with duplicating existing software products, or specific features within them. Simple command-line tools that implement obscure hashing and encryption algorithms are straightforward initial targets, but this approach can easily extend to more complex software, such as websites, professional software, and games."

One of our best reports yet @delphi_intel from Can. Browsers will likely prove the most important real estate in AI. This report covers the industry's evolution, the value chain, and the current war for what is likely one of the most valuable data sets in AI. High ROI read.

The app-centric way of interacting with computers is dying When the iPhone arrived, it triggered the big bang of apps. We've spent the last fifteen years training ourselves to think in apps. Everything begins with an app: unlock your phone, hunt for the right icon, tap, scroll, repeat. But if you pause for a second, it feels wrong. Our brains don't naturally organize life into apps; we think in terms of intent. Before Large Language Models (LLMs), apps were the best abstraction we had to translate our intent into actions. And, to be fair, it worked perfectly… until now. Then ChatGPT happened, and suddenly, hunting through endless screens of icons feels like busywork from another era. Now that computers have learned to talk, we should be able to tell them what we want, and they should take care of the rest. The next decade is all about closing the gap between where we are today and this ideal state. The immediate steps are quite clear. Largely there are 3 gaps that most people seem to agree on: AI isn’t personal enough – Despite all its magic, it still knows very little about you. AI isn’t ubiquitous enough – It doesn’t meet you where you are. AI isn’t proactive enough – It tells you what to do, but more often than not doesn’t do it for you. So, where will this transformation happen? Which platform will a more personalized, ubiquitous, and proactive AI take root in? It's hard to predict the distant future, but the ideal immediate candidate is already right in front of us. It's not a new app, a wearable pendant, or a futuristic interface. It's the software we've been using all along: the browser. There are a few reasons why the browser is uniquely positioned: First is the context layer. Tabs, domains, sessions, iframes, cookies, credentials, extensions, bookmarks, history, sidebars, URLs—these are all rich sources of context through which browsers continuously gather data. They observe how you navigate the web, what you search for, what you ignore, and what you revisit. Your default browser already holds deep context about you. Next, a lot of work happens in the browser. Much like the iPhone outgrew the “phone” label, the browser has outgrown “browsing.” We carry most of our knowledge and productivity work—email, docs, calendars, meetings—on the web. Modern browsers have become quasi-operating systems, hosting not just static content but full-fledged apps—Google Workspace (Docs/Sheets/Slides/Drive), Slack, Figma, Notion, Asana, PWAs, crypto wallets—the list goes on. With AI, the browser is ready to host a new kind of software: intelligent agents capable of chat, coding, deep research, computer usage, image generation, and more. As it leverages these agents, the browser is no longer just where work happens but becomes an active participant in that work. We won’t just use apps in the browser, browsers will become an "embedded" part of our lives. Finally, the browser is cloud-native and therefore cross-platform. iOS, Android, macOS, Windows, iPadOS—even TVs and VR headsets—the browser runs everywhere. Agents within the browser would feel ever-present, offering a unique quality that standalone AI chatbots can't match. All of this highlights the motivations behind the looming Browser War III—a fight to control the web’s most critical gateway in the age of AI. Browser War III The Browser War might be the highest-stakes war of all because it's a winner-take-all market. Browsers are extremely sticky. The longer you use a browser, the more footprint you leave—profiles, passwords, credentials, bookmarks, extensions—and the less likely you are to switch. With AI agents capturing your valuable web behavior data and offering hyper-personalized services, this stickiness will only deepen. But this intense market concentration isn't driven by consumer choice alone. The browser market is tightly controlled, with operating systems and OEMs (Original Equipment Manufacturers)—upstream of browsers—playing a significant role. iOS and Android intentionally design their interfaces to lock users into preferred browsers Safari and Chrome, through measures like restricting browser engines, obscuring default browser settings, and nudging users toward defaults. OEMs further reinforce these walled gardens by pre-installs (like Edge on Windows PCs or Chrome on Android devices), often bound legally or incentivized financially to do so. Under normal conditions, there would be no chance for a new winner to emerge in such a controlled environment, which explains the historical lack of competition. But we live in unprecedented times. No incumbent—browser, OEM, or OS—is safe amid such disruptive technology. AI browsers offer entirely new experiences, blending chatbot interfaces directly into browsing. Navigating product design and engineering breakthroughs in this space is part science, part art. Those who succeed in doing so now have the chance to build lasting moats for decades. Within this evolving landscape, several emerging players are making their mark, including Perplexity Comet, The Browser Company’s Dia, Genspark, Fellou, Opera Neon, and the much-speculated browser from OpenAI. However, before delving into these new entrants, it's crucial to first grasp the current state of affairs. Game of Thrones Below is web traffic flow measured in page views across key players—ranging from OEMs and operating systems upstream, to search engines downstream of browsers. Here are a few highlights from the data:

Remain thankful and learn to become unborable, and you have basically already won

Been exploring early-stage robotics companies recently. We've all heard of Figure and other big names, but watching what's happening at grassroots level is super interesting. Here are 11 companies founded since 2023 with < $100m in funding that caught my eye. Thread: 1/17.