7 posts from the last 24 hours on temperature2.

White House finalizes AI review framework, keeps it secret

The White House says it met its deadline for a voluntary AI cybersecurity review framework ordered by Trump in June, but won't disclose the contents, who's seen it, or when labs start using it.

Horizon3 triples to $2B valuation on AI-vs-AI bet

Horizon3.ai raised a $250M Series E at a $2B+ valuation, tripling in 13 months on 120% ARR growth from its autonomous pentesting platform NodeZero.

Uzbekistan, Kazakhstan race to build Central Asia's AI hubs

Nikkei Asia reports Saudi-backed DataVolt and an Nvidia-linked Kazakh campus are both racing toward 2026-2027 completion, turning the region into new AI infrastructure territory.

Alibaba's Qwen3.8-Max launches with 2.4T parameters

Alibaba's new flagship model claims second place behind Claude Fable 5, with open weights due next week and a workplace-agent platform launched alongside it.

Why an LLM can know the truth and still get it wrong

Alibaba and Zhejiang University researchers name the CHOKE phenomenon: models whose internal representations know the right answer but output the wrong one anyway.

A single A10G GPU now serves Gemma-4 at 510 TPS

A six-day Hugging Face and Google challenge to speed up Gemma-4 inference on one A10G GPU ended with a fully open recipe hitting 510 tokens per second.

How YaRN Stretches RoPE Past Its Training Length

Qwen3.6 trains natively at 262K tokens and stretches to 1M with a rotary-embedding trick called YaRN, not a bigger model. Here's how compressing position math without retraining actually works.

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