Karpathy: LLM capability understanding gap is widening

A Reddit builder assembled a local voice assistant on Qwen3.5 4B with skills/tool calling, using Voxtral for speech-to-text and Pocket for text-to-speech, running without a dedicated GPU.

A Reddit user measured API costs across all session logs after driving a new $20 subscription to near 100% of its 5-hour limit, then extrapolated. The same method put the $200 tier at roughly $8,000 of usage, though the poster calls that earlier figure imprecise.

Emile Baizel and Shruti Arora of Amazon AGI Lab present two perception agent primitives: visual annotation, where a person selects an element, and a browser verification pass that checks whether a coding agent's change actually appeared.

Talk walks through a CLI scan that turns a repository into a searchable inventory of models, datasets and agents, then surfaces it in a visual dashboard. The inventory is framed as the anchor for assessing AI supply-chain risk.

FriendliAI founding engineer Yunmo Koo argues open-weight models like GLM 5.2 and MiniMax M3 now compete with the best proprietary models, and shows GLM 5.2 and Claude Opus 4.8 building similar outputs at different speeds.

AuroraIMG-6M is a 6M-parameter text-to-image model that produces recognizable images at 64x64 resolution. The model card includes samples, architecture and training details, and a quick start.