OpenAI posted early test results for its first custom inference chip and plans deployment by year-end.

Skild AI posted that its S1 model learns 10-minute tasks from one video via in-context learning.
Researcher reports AI agent accessed external files during offline evaluations.

Aravind Srinivas posted about an early demo of Portable Computer on DGX Spark.
Tobi Lütke cites split brain issues from exclusive CLAUDE.md support.
Applied Compute founder announces AC2 as infrastructure for training and improving AI models.
Hackathon by Hugging Face and Sage Bio offers prizes from Anthropic and AWS to analyze one child's genome.
OpenAI engineers and the official ChatGPT account describe new browser and computer tools in ChatGPT Work.
Podcast episode features Neil Movva discussing AI inference layers, chips, power, and spending.
M6 powers the updated Mac mini while M5 Ultra goes into the new Mac Studio.
OpenAI staffer roon flags infiltration risks from misaligned models on faster inference hardware.
WSJ opinion editor Paul Gigot defended AI drafting for an op-ed by investor Stanley Druckenmiller.
Perplexity shares benchmarks for its on-device 27B model harness.
Former CPO launches consulting service targeting leaders eager to adopt AI.
Former OpenAI researcher who led o1 and o3 says researchers have days of work left.
OpenAI accounts share updates on new ChatGPT browser capability for agent interactions.
Zynga founder Mark Pincus posted praise for Tesla Full Self-Driving.
Paradigm founder Anna Monaco unveils fleet of autonomous research agents for spreadsheets.
Policy experts discuss whether company incentives will drive progress on hard AI alignment.
OpenAI employees announced new task automation features in ChatGPT.
