Martynas Jusevičius

Extraordinary that Trump's allies in Europe are three dictatorships, Russia, Hungary and Belarus What he loathes is any kind of democracy

June 1983. A 28-year-old Steve Jobs walks into a design conference in Aspen, Colorado. He asks the room who owns a personal computer. Nobody raises their hand. He says “Uh-oh.” Then he spends the next 55 minutes describing the next four decades of technology. Jobs told the audience Apple’s strategy was to “put an incredibly great computer in a book that you can carry around with you, that you can learn how to use in 20 minutes… with a radio link in it so you don’t have to hook up to anything.” That’s an iPhone. In 1983. The Mac hadn’t even shipped yet. He described an MIT project that sent a camera truck down every street in Aspen, photographed every intersection, and built a virtual walkthrough on a computer screen. Google Street View launched 24 years later. He said office networking was about 5 years away and home networking 10 to 15 years out. The web went mainstream in the mid-90s, about 12 years later. Dead on. He described software being sent electronically over phone lines, with free previews and credit card payment. That’s the App Store, 25 years before it launched. He even compared it to the music industry and said software needed “the equivalent of a radio station” for free sampling. Apple built the iTunes Music Store 20 years later. The AI prediction is the one that hits different now. Near the end, Jobs talked about machines that could capture a person’s “underlying spirit” or “way of looking at the world,” so that after they died, you could ask the machine questions and maybe get answers. He said 50 to 100 years. ChatGPT arrived in about 40. The weird part is this speech was lost for nearly 30 years. The full hour-long recording only surfaced in 2012 when a blogger got a cassette tape from someone who attended the original conference. The Steve Jobs Archive didn’t release actual video footage until July 2024. His timelines were consistently too fast. He wanted the “computer in a book” within the 1980s. Apple’s first attempt was the Macintosh Portable in 1989, which weighed 16 pounds and cost $6,500. The iPad arrived in 2010, 27 years late. He guessed voice recognition was about a decade away. Siri launched in 2011, nearly 30 years later. The vision was right every time. The clock was wrong every time. Apple was doing about $1 billion a year in revenue when Jobs gave this talk, with under 5,000 employees. Today it’s worth $3.7 trillion.

Dug into /autoresearch and wrote about how it works for dummies like myself, it's insanely cool @karpathy justinluk.medium.com/karpathys-630-…

@WhiteHouse “This was the sixth reich” youtu.be/nX3WgrQ2uVY

STRIKE. 💥🦅

@AusiasTsel Nobody is using SPARQL on Wikidata? :)

The biggest surprise was Wikidata. It has structured translation data for thousands of works; editions exist in which languages, linked to original works. But it’s locked behind SPARQL, script-matching across alphabets, and author alias resolution. Nobody was using it for this.

I spent 4 months trying to answer a simple question: has this book been translated into my language? Turns out no one tracks this. Not ISBN. Not Amazon. Not Google. Not libraries. So I built a tool that crosses four databases to piece it together. zenodot.app

@levelsio Except APIs are a dumb legacy paradigm 🤷‍♂️

Thank god MCP is dead Just as useless of an idea as LLMs.txt was It's all dumb abstractions that AI doesn't need because AI's are as smart as humans so they can just use what was already there which is APIs

Three sources tell 60 Minutes that undercover agents purchased a miniaturized microwave weapon from a Russian criminal network. Secret U.S. military lab testing of the device on rats and sheep has resulted in symptoms similar to Havana Syndrome, a confidential source says. cbsn.ws/4d7bHkz

@seralf @karpathy I have adapters that virtualise REST APIs such as GitHub as SPARQL endpoints :) Meaning that semantic queries can be executed on the fly, without having to materialise the data first. Works better for lookups than aggregations due to the execution constraints.

@karpathy Have you consider the idea of describing each repository by metadata RDF, than having each agent starts by navigating over the reachable part of the emerging text-graph? This way agents should look / "read" only the promising paths, and some network analysis may be used as well.

The next step for autoresearch is that it has to be asynchronously massively collaborative for agents (think: SETI@home style). The goal is not to emulate a single PhD student, it's to emulate a research community of them. Current code synchronously grows a single thread of commits in a particular research direction. But the original repo is more of a seed, from which could sprout commits contributed by agents on all kinds of different research directions or for different compute platforms. Git(Hub) is *almost* but not really suited for this. It has a softly built in assumption of one "master" branch, which temporarily forks off into PRs just to merge back a bit later. I tried to prototype something super lightweight that could have a flavor of this, e.g. just a Discussion, written by my agent as a summary of its overnight run: github.com/karpathy/autor… Alternatively, a PR has the benefit of exact commits: github.com/karpathy/autor… but you'd never want to actually merge it... You'd just want to "adopt" and accumulate branches of commits. But even in this lightweight way, you could ask your agent to first read the Discussions/PRs using GitHub CLI for inspiration, and after its research is done, contribute a little "paper" of findings back. I'm not actually exactly sure what this should look like, but it's a big idea that is more general than just the autoresearch repo specifically. Agents can in principle easily juggle and collaborate on thousands of commits across arbitrary branch structures. Existing abstractions will accumulate stress as intelligence, attention and tenacity cease to be bottlenecks.

How to Build Knowledge Graphs That Enable AI-Driven Enterprise Applications by @Gartner_inc gartner.com/doc/reprints?i…

Meta just showed you can build billion-parameter foundation models for graphs 🕸️ GraphBFF brings LLM-style scaling laws to graph data — bigger model or more data = predictably better results. arxiv.org/pdf/2602.04768

Sam Kriss (@samkrissfeed) reports from San Francisco on the next generation of AI startups and their “highly agentic” founders. harpers.org/archive/2026/0…

@windley @VaughnVernon Try to cross-reference data across multiple datasourcea/services

@VaughnVernon Yeah, I can see how it might help, but I also think it might be overkill in a world where software understands intent. Not that LLMs are perfect at that, but they really cut down on the need to be as explicit about everything.

I've been wondering whether the semantic web makes more sense now and possibly even be in demand. /cc @windley

@bebraw @udaysy @jdegoes linkedin.com/posts/martynas…

@udaysy @jdegoes Now with agents, you have all the reason to make the web semantic. Also see researchgate.net/publication/40… for the bigger picture.

All new software and even websites should be built from this assumption: The primary user is going to be an AI.

@soohoonchoi x.com/namedgraph/sta…

why hasn't the semantic web taken off yet? esp with all of this recent talk about semantic, knowledge, skill graphs? the rfc's for all this exist.

@joshu @soohoonchoi

@soohoonchoi because semantic web is mostly about the semantics of the schema, not the semantics of the data

@spenley Why did Solid fail? linkedin.com/posts/martynas…

Tomorrow I interview Sir Tim Berners-Lee on-stage at Talent Arena. I have a million questions for him already, of varying levels of geekdom. I even think I may have one question that he's never been asked before. But what interesting question do you want asked, that he might never have been asked before?

@IAPonomarenko @grok does @rogerboyes have a history of failed predictions? Any specific to the Ukraine war?

How many more times do these people need to get hit in the face by the reality before their publishers finally begin noticing?