AI Briefing#open-source-models
The latest AI and developer news about #open-source-models, with the original source and a short summary.
The latest AI and developer news about #open-source-models, with the original source and a short summary.
Researchers demonstrate that removing a single vector from the residual stream of 13 open-source models up to 72B parameters effectively disables refusal mechanisms.
The system uses an Experience Flywheel and staged learning to close the loop between experience production and policy improvement.
Executives are increasingly adopting sovereign AI strategies to protect proprietary data from potential leakage or competitive use by frontier AI providers.
The new Holo4 models, available in 27B dense and 35B-A3B MoE sizes, interact with software across GUI, code, MCP, and API interfaces.
Together AI has released tev1-4B-experimental, a Jev-like classifier fine-tuned on Qwen3.5 4B, available on Together serverless for $0.042/M input tokens and $0/M output tokens, with a reported training cost of just $17.
Pirate Face, a decentralized infrastructure that mirrors Hugging Face's open models via torrents to protect them from deletion risks, has been released.
Critics argue that ArtificialAnalysis's LLM intelligence-cost graph distorts actual cost differences due to its use of a logarithmic scale and unrealistic pricing.
Kimi Vendor Verifier validates implementation deviations across inference providers.
Even small open models succeeded in detecting zero-day vulnerabilities and reproducing exploits.