AI1 min read
Dario Amodei proposed embedding third-party evaluators inside Anthropic. Sam Altman confirmed OpenAI would commit to the same practice.
Why it matters: Independent access to training data helps verify if models hide unsafe behavior during testing.
By OpenSmartRoute editorial · written through the router by writer-small
From TechCrunch AI - “Anthropic and OpenAI want to embed safety evaluators. Will they really be independent?”
Anthropic CEO Dario Amodei published a proposal for embedded safety evaluators.
He wants independent groups to report incidents and assess model alignment.
OpenAI CEO Sam Altman agreed to commit to this practice.
Evaluators like METR and Redwood Research would get unprecedented system access.
They could check intermediate training versions called checkpoints instead of just finished models.
This helps find how models behave during training, not just when released.
Keep reading
AI1 min read
Knowledgator released GLiFormer, an encoder that extracts nested JSON without generating tokens.
AI1 min read
Anthropic CEO Dario Amodei wants outside groups to audit AI safety. Experts say fixing network security basics is more effective.
Anthropic combines its chat and Cowork tools into a single window. This removes the need to switch tabs for different tasks.
