Andrej Karpathy

Tencent used TeamAI-CLI internally since March 2026, then open-sourced it on September 7. It puts a team's skills, rules, and docs in one git repo, merges land on everyone's next agent session automatically, and each learning earns a confidence score from real usage. Here's what shipped and how it compares to other team-memory tools.

Recurrent depth is a technique where a model reuses a block of layers multiple times on the same hidden state instead of externalizing extra reasoning as chain-of-thought text. It's efficient and it works — but it moves reasoning into a space nobody can read, which is exactly the concern researchers raised about OpenAI's Astra.

Y Combinator's Garry Tan published new evals for GBrain, his open-source agent-memory layer, on August 31, 2026 — claiming SOTA retrieval without an LLM in the loop. One reply asks the question the whole post is built around: the eval set and the retrieval layer share an author, so what does GBrain score on a benchmark nobody else wrote?

img2threejs v1.5.1 ships the Character Update: a 2D reference image becomes a rigged, animated Three.js fighter with measured punch VFX, sweat particles, and 19 combat clips — pure TypeScript in the browser, no Blender workflow. explainx.ai breaks down the GLB-measurement pipeline, how it differs from neural mesh tools, and what web-game builders should steal.

TIME published the 2026 TIME100 AI on August 27 and the replies filled with the same complaint: no Jensen Huang, no Demis Hassabis, no Karpathy, but Paris Hilton and Ben Affleck made it. Read the list carefully and the pattern is deliberate — Nvidia's seat went to its head of sustainability. explainx.ai breaks down what TIME actually measured and which of your daily tools have their makers on the list.