5 posts from the last 24 hours on temperature2.

Meta's Muse Spark 1.3 cuts tokens, keeps 1.2's price

Meta's new coding model needs 25% fewer tokens per task than 1.2, but keeps the exact same per-token price.

Gemini 3.8 Flash ships with a government-only cyber twin

Google DeepMind shipped Gemini 3.8 Flash on September 2, 2026, alongside a restricted vulnerability-patching variant gated to governments and infrastructure operators.

Trump administration backs OpenAI in NYT copyright fight

The White House told a Manhattan federal court on Sept 2 that training AI on copyrighted text is fair use, the first time it has intervened in a major-publisher AI copyright suit.

Signals: AI tutors, cheaper serving, and looped MoEs

Microsoft Research's StudentSim out-simulates GPT-5.4 as an AI tutoring proxy, a self-hosted LLM consolidation beats a 7x larger baseline, and ByteDance Seed trims pretraining FLOPs with looped MoE layers.

AMD beats Nvidia for the EU's 450 million dollar AI supercomputer

EuroHPC JU signed a 387.8 million euro contract on August 31, 2026 to build LUMI-AI on AMD Instinct MI430X chips instead of Nvidia's.

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

The numbers first, all day: follow temperature2 on X.