Crabhaus

What matters in AI, and why.

← Today's edition

Full wire feed

Sun 4 Oct · 8 stories · 4 sections

Frontier labs 1 item

Aleph Alpha releases Kolibri, an English-German MoE model (78B total, 3B active) with 1M-token context, Apache 2.0 on Hugging Face.

Agents & tooling 4 items

Simon Willison: pay-per-use AI APIs need default hard budget caps; warning emails won't cut it once coding agents spend on your behalf.
datalab's Lift, a 9B open-weights model, extracts schema-matching JSON from PDFs in one pass and returns null instead of guessing.
Anthropic's guide to Opus 5.5 in Claude Code: the model works longer on its own, so define what 'done' looks like and when to stop.
Essay: vector-snippet 'memory' misses what agents need — documentation of where things live, why they were built, and what was agreed.

Research & papers 1 item

Microsoft and HF's ThinkingBox grades agents on the backend state they leave behind, re-running tasks to score reliability, not prose.

Business & policy 2 items

An OpenAI safety leader quit, calling the company's culture 'broken'; he wrote the safety reports for its major model releases.
A US federal judge ruled a warrantless Flock license-plate search unconstitutional: 'indiscriminate mass surveillance.'
Back to top ↑