What does personal intelligence mean? If tech like Palantir’s were to be democratized and available to you, what might your own personal seeing stone in your pocket enable you to know, to do? Here’s a bit of a sneak peek through ASIMOV’s seeing stone regarding our San Francisco event later today. In the minute or so that this ingest took, we first extracted the Luma event’s guest list and then for each of the 95 guests figured out their social media accounts on X, Instagram, LinkedIn, and beyond We obtained those profiles in full, as well as the respective social graphs (followers/followees/mutuals) for each of those accounts, and collated all that into a nontrivial (95 → ~300K → 50M+ nodes) knowledge graph representing the combined social network, reconciled with and layered onto our existing massive knowledge base. This tells us who is in the room and how they relate to us and to each other We then fetched and indexed the most recent posts for each guest’s social media accounts, and computed & compiled executive summaries of their recent activities and demonstrated interests on each those platforms, not only from text posts but also from any available photos and (short) videos. These dossiers are incredibly helpful in pinpointing who might share mutual interests We also, naturally enough, indexed all faces detected in guests’ profile photos and their most recent social media posts, as well as filed away a voice sample from any available Instagram or YouTube clip. This enables recognizing guests by their likeness and/or their voice, as we have showcased at previous events All this, by the way, can run directly on your phone, tablet, or laptop as well; just rather slower than the dedicated, scalable upstream infrastructure showcased here. Nonetheless, in the future you will be able to produce and monetize this event intelligence in your pocket yourself, or just purchase on-the-go access to it as you like The future is already here, it's just not evenly distributed. Join us at Frontier Tower tonight at 6pm to learn & discuss personal intelligence built on your own personal context graph connecting all people and information in your life!

NEW: According to a bombshell report in the New York Times, Anthropic co-founder Chris Olah threatened to walk out of Pope Leo XIV’s AI encyclical launch in May because the pope rejected the idea that machines can be conscious. Olah’s team then privately lobbied the pope’s advisers “to take the possibility of model consciousness seriously.” Pope Leo XIV held firm. For months, Anthropic has wined and dined theologians and religious scholars under nondisclosure agreements, hoping they would bless the idea that Claude has moral standing. thelettersfromleo.com/p/nyt-…

NB: Popes have been talking about ontology for two millennia.

Introducing Underdog, your Private Personal AI on devices you already own Today we're announcing our backing from @a16z @khoslaventures @HummingbirdVC Anthology (@AnthropicAI @MenloVentures) @patrickc @naval @rauchg @polynoamial @Thom_Wolf @Mascobot @tszzl @OfficialLoganK and other top AI leaders Our mission is to provide free, capable, reliable and private AI to billions of people. Underdog’s Law: today’s frontier intelligence reaches your devices in six months. We’re starting with fast, capable AI that runs on consumer hardware. By co-designing models and inference engines, we’re pushing the frontier of capability, speed, power and data efficiency. Our research also spans agentic commerce, confidential inference, and how AI will reshape the internet economy. Privacy and capability no longer needs to be a tradeoff. If you believe in this future, join us. Time to build.

If AI agents spends more time waiting than generating answers, what should we actually be accelerating? And what does that mean for the next generation of AI chips? A look at where the time, memory and money go when agents get to work.

Turns out raw GPU token speed isn't the real bottleneck for agents. It’s CPU tool execution, memory bandwidth, and sandbox overhead. Interesting insights from Kevin.

A couple of useful websites to bookmark from these comments: resetbeacon.com willreset.com

So proud of the team for this one. Same near​.com, completely reimagined. My favorite part, press ⌘K and type “tesla”. The stock and the perp, side by side.

The promised ChatGPT global reset is late, and the comments & proposed community note are hilarious!

It's hard to overstate my distaste for OpenSSL popping up in the transitive dependency graph OpenSSL is the achilles heel of otherwise clean, safe cross-platform Rust builds, and must be ruthlessly exorcised whenever it returns to haunt us The collective time the Rust community has wasted troubleshooting OpenSSL builds this past decade and more would ship Rustls a thousand times over

We actually went deeper on this here:

Herding a dozen Astra (max) agents with @herdrdev to usefully burn remaining compute quota before the global reset. Somewhat perverse incentive structures in this new age, at least for the token-poor

Phenomenal!

That's just badass. Well done, @Figure_robot!

It's becoming so evident now that Rust is going to sweep the systems programming space at the expense of just about everything else--and perhaps not only systems programming... Coders I know who worked the previous decade with Python, Ruby, or TypeScript have over the past quarter switched to vibecoding Rust. Why not? Faster, safer, smaller (memory-wise), simpler (deployment-wise), what's not to like? Similarly, Rust is eating the Python ecosystem from the inside outwards (uv, Ruff, Polars, Pydantic, etc, etc) Rust is an emerging Schelling point for coding agents. Its infamously strict static constraints function like guide rails, such that by the time the code actually compiles, the job's often largely done. And Rust's best-in-class, precise compiler error messages (valuable when compile cycles are slow) help agents make quick work of pinpointing & troubleshooting the effects of their changes We didn't yet see this sweep prior to this year because models were simply not capable enough of fully mastering the nontrivial complexity of the language (cf. ownership). They still made mistakes that I now simply no longer see with the current frontier. Rust seems to have a particularly pronounced threshold effect: the same strictness that repeatedly blocked a weaker model now actually helps the stronger model converge! For my own work, Fable 5 was the inflection point where the quality of generated code often exceeded what would be economical to code manually. (Astra and Opus now mostly do okay, too) This new Rust hegemony will become obvious to everyone during 2027, with much gnashing of teeth!

Speed is what every emulator needs. What is your favourite Need for Speed version? Which one do you want to play on your phone?

Aviation solved AI slop in 1986. Karpathy is telling people to ask their LLM to write in ASD-STE100, a controlled language built for aircraft maintenance manuals. I went down the rabbit hole on where this spec comes from, and the backstory makes the hack even better. Back in the late 1970s, European airlines were handing English manuals to mechanics around the world who mostly didn't speak English as a first language. One ambiguous sentence in a fuel line procedure can get people killed. So the industry built a version of English where ambiguity is banned at the dictionary level. The spec allows roughly 900 approved words. Each word gets one meaning and one part of speech. "Close" can only be a verb, because "close the door" and "the close door" can't be allowed to coexist in a repair manual. "Follow" only means to come after. If you want obey, you write obey. Procedural sentences max out at 20 words. One instruction per sentence. Active voice only. Around 1,200 common words are explicitly banned, each with a mandated replacement. You don't "commence pumping." You start the pump. English has about 170,000 words in current use. Aviation decided mechanics get 900, and planes became safer to maintain because of it. Now flip it to LLMs. Models hedge and synonym-cycle because the training data rewards sounding fluent. STE is 40 years of accumulated rules for stripping exactly that out, written by people whose readers would die if a sentence could be read two ways. It might be the strongest anti-slop prompt in existence, and it's a free PDF.

I think we have a problem

People of Pi: We've shipped Pi 1.0 with Pi Durable. Go make them yours. earendil.com/posts/pi-1-0/

I think you can go even more low level when humans don't have to write and read code: watx.berrry.app this is what I'm using to build Windows emulator memory management was definitely a huge issue without extra guardrails