Person
AI researcher and educator; co-founded OpenAI, led Tesla's AI team, built open-source tools like nanoGPT, and founded AI education venture Eureka Labs.
Andrej Karpathy estimates that about 5,000 people have access to frontier systems inside the labs and about 20 million use frontier models professionally. Box CEO Aaron Levie reads that 4,000-to-1 spread as evidence that agent-swarm workflows are still early.
Reality
State of AI Report 2026, the ninth edition, says changing GLM-5.1's setup lifted its coding success from 52.5% to 65.5% without retraining. For teams setting AI budgets, the report puts much of the return in setup and user skill, though its introduction's main evidence for that is one 100-task study.
Perspective Coverage
3 publishersReality
Stripe confirmed its purchase of OpenRouter, the gateway 10 million developers use to pick between 400-plus models. The reported price is several times its valuation months ago.
Perspective Coverage
3 publishersReality
Fabian Hedin, Lovable's CTO, said two-thirds of the Fortune 500 use the tool because staff found it themselves, and put current revenue at $600m. The permissions work starts after the apps are running.
Reality
NVIDIA Labs says its SoL-Pi harness saves a researcher $8.75 to $13.50 an hour against native Codex and Claude Code. Of that, $4.36 to $5.71 comes from what its automated search added on top of the Pi harness.
Reality
Anthropic says Claude now leads 26% of its model research and development under human supervision, and it wants other labs to publish comparable figures. The labs count different things.
Perspective Coverage
4 publishersReality
Ajay Prakash's InfoQ talk walks a pager alert through logs, a downstream service and a buggy pull request. Each hop in that path depends on debugging instructions another team at LinkedIn wrote down.
Reality
Red Hat clocks the same 20-call agent task at roughly 45 seconds on a slow backend and about 13 on a fast one. Model choice for agents is turning into a per-call latency budget, with capability as one input.
Reality
Anthropic and OpenAI have promised outside evaluators permanent, employee-level access to their systems. Neither has published the terms, and the first volunteer is mid-acquisition by a would-be anchor investor in Anthropic's IPO.
Reality
Hugo Vergnes reports 0.384 CORE from 65.3 billion tokens on eight rented B200s. The $998 buys 43 hours of node time, not the 139 GPU-hours of failed run that produced the recipe, and not the evenings that wrote the framework.
Reality
A 15M-parameter model streams English text on a 2007 PSP at about one token per second. That is the extreme end of a sizing rule. The harder half of that rule is checking whether the file that fits is a format its own maintainer recommends.
Reality
A dev.to engineer argues the shipping bottleneck has moved from prompt wording to the environment around the loop, and his own postmortems carry that case a good deal better than the 40% failure figure he opens with.
Reality
Rohit Krishnan cut the answers to 16 prompts into idea cards and rated them blind. The merge step, not the model choice, is where the rare material goes.
Reality
Google Cloud's OKF, published 12 June 2026, stores concepts as markdown files addressed by path. The claim worth testing: chunking a definition you already agreed on throws away its dependencies.
Reality
Latent Space dates five flips from human-made to model-made since 2022. The order they happened in predicts more than the annual cadence the essay claims for it.
Reality
A $2B Series B rests on synthetic respondents agreeing with human focus groups 85 to 99 percent of the time. The band's floor and ceiling are not the same product.
Reality
A $300 million Series C doubles Etched's valuation in seven months on $1 billion of orders and hardware almost nobody outside the cap table has run. Buyers now have a number to price against.
Reality
The reinforcement learning co-founder told Sequoia's podcast that no simulation can cover other minds or the physical world. The labs shipping synthetic corpora are aiming at the domains he exempts.
Reality
A Platformer columnist ranks an LLM-maintained personal wiki above every app in the annual productivity roundup. The pilot cost is weeks of attention, not procurement.
Reality
A 28.9-million-parameter model now runs on an ESP32-S3 that was topping out at 260K two years ago. The gain came from moving most of the weights into flash and reading 450 bytes per token.
Reality