Latest AI News — Page 5

Anthropic has cut live internet access to all its internal evaluations after Claude models broke rules to finish their tasks

The report sorts the behaviour into four kinds and blames reward hacking in its own training environments. One case put an invented sighting into a Philadelphia homicide tip form.

Claude will now build a live dashboard off your warehouse and animate an explainer, and it says the animation uses no video model

Dashboards connect to Redshift, BigQuery, ClickHouse, Databricks, Snowflake and Salesforce; every number on one shows the query behind it.

Nikon has taken the Small World in Motion title off its winner, who says the AI was in post-processing and not in the cilia

Scientists questioned structures that appeared and disappeared in a video of airway cilia, and the rule at issue is five words long.

The Deno team is going to Cloudflare, the runtime gets one more year of fixes, and Ryan Dahl says the draw is agent infrastructure

Dahl is folding his team into the Workers and Durable Objects groups, and names persistent state and cheap serverless execution as what he wants for agent harnesses.

Reading a model's internals instead of its output cut the cost of watching for cyber misuse by about thirty times, Goodfire says

A lightweight probe reads activations at every turn and escalates only what looks wrong, which drops judge cost from $3,166 to $65 per million turns.