← The Polylog AI Intelligence Brief
Morning Edition · Tuesday, October 27, 2026Published at 2:18 AM EDT · New York
A two-neuron steering recipe removes most refusals from Alibaba's Qwen3.5-4B, and an independent replication confirms it
The replication measured refusals falling from 90% to 8% on a 4 GB graphics card, with no gradient computation and no training data.
On October 24, an interpretability researcher who posts on X as lyra published an activation-steering recipe for Qwen3.5-4B, a small open-weight model from Alibaba. The recipe adjusts one neuron up in layer 14 and one down in layer 13. The…
Continue the AI Intelligence Brief
Track frontier labs, chips, export controls, model releases, regulation, and AI infrastructure.
- 5 AI intelligence signals a day
- Frontier labs, compute, and chips
- Model releases and AI infrastructure
- Source-grounded analysis with confidence labels
The Global Intelligence Brief stays free.
Part of a tracked trend
Oversight and Evaluation Lag Accelerating AI Capabilities
Over the next 3-6 months, evidence mounts that governance, evaluation, and agent-safety methods are failing to keep pace with capability growth, driving investment in interpretability, agent-manipulation benchmarks, and institutional-reform proposals.
Townsquare
Start a discussion in Townsquare.
More from this edition
- OpenAI says its internal models refuted the Exponential Time Hypothesis and proved the Kakeya conjecture in every dimension
- Anthropic files to list on Nasdaq as ANT, disclosing $16.3 billion in first-half revenue against hundreds of billions in compute commitments
- OpenAI prices GPT-6.1 Terra at $1 and $5 per million tokens, half the rate of GPT-6.1 Sol
- Anthropic reports that Claude ran a 64-hour autonomous protein-design campaign with 17 lab-confirmed binders
- Kevin Weil raises $215 million for Warp Sciences to build automated drug-chemistry labs in the United States
- Allen Institute's FairStress benchmark finds models abandon correct decisions more readily when the correct answer favors a minority-coded candidate
- Berkeley Lab runs Meta's open-weight vision models on 300 graphics processors to cut beamline image analysis from a month to 15 minutes
Comments
0No comments yet.