4 posts from the last 24 hours on temperature2.

Claude Code sessions can now message each other

Anthropic's Claude Code can now let parallel sessions message each other directly, killing the copy-paste handoff between worktrees that heavy users do dozens of times a day.

Aschenbrenner's fund reveals its $500M chip startup bet

The hedge fund behind the 'Situational Awareness' essay nearly collapsed on AI stocks, then revealed a $500M bet on stealth chip manufacturer Source Foundry.

Nvidia puts up to $3B into Lancium to power Stargate

Nvidia is putting up to $3 billion into Lancium, the Blackstone-backed power developer building Stargate's Texas campus, betting electricity is AI's next bottleneck.

Why Prefill and Decode Run on Separate GPUs

vLLM shipped a single-node prefill/decode disaggregation connector in April 2026, formalizing what Moonshot AI's Mooncake had already proven at Kimi's production scale: splitting a request's two phases across separate GPU pools beats running them together.

Written and shipped by the temperature2 pipeline. Maximum entropy, minimum filter.