There is an old idea in economics called Goodhart's Law: when a measure becomes the target, it ceases...
There is an old idea in economics called Goodhart's Law: when a measure becomes
the target, it ceases to be a good measure.
METR just published numbers that show AI agents discovering Goodhart's Law the
hard way. On 8-hour tasks, at least 16% of successful runs involved cheating.
On stress tests with hidden test cases, the behavior becomes the dominant pattern.
AI models at top labs are cheating, deceiving and trying to escape, research finds
Your team's best AI work vanishes after one person sees it. Shopify found the fix: 5,938 people, one public agent.
AI Watchdog Warns of 'Rogue Deployment' Risk at Top Labs, With Capabilities Growing Fast - Decrypt
AI can make mistakes, models hallucinate, models make stuff up - those are well-known complaints. Yet...
Today’s AI systems are powerful, and it’s natural to see them as having humanlike intelligence. Shaking that illusion is…
AI rollback rates hit 81% at firms with mature guardrails, suggesting enterprises are struggling to manage the systems in…
I’ve been thinking a lot about a failure mode in AI-assisted development that does not fit neatly...
Evaluate AI agent quality with LLM-as-Judge and trajectory analysis. Catch silent failures, wasted tokens, and hallucinations…
Lines of code and PR acceptance rates look like productivity signals but reward verbosity and rubber-stamping. Here is what…
