Mastodawn

The senior director of AI at AMD has complained on Github that since January, Claude now ignores instructions, suggests fixes that are incorrect, directly contradicts requests and claims completion despite, well, not completing the task.

A new software engineering hazard is your developer velocity getting worse because your AI agent provider is going through capacity issues.

https://www.pcgamer.com/software/ai/amds-senior-director-of-ai-thinks-claude-has-regressed-and-that-it-cannot-be-trusted-to-perform-complex-engineering/

AMD's senior director of AI thinks 'Claude has regressed' and that it 'cannot be trusted to perform complex engineering'

This is reportedly since changes made in February.

PC Gamer

Show thread

Kees de Kooter 🍉13h ago

@carnage4life as if it was trustworthy in the past

Show thread

Phil Wolff 12h ago

@carnage4life when your employees have troubles outside your workspace that affect their work... more closely checking _all_ their work. coaching, counselling, supportive services, leave. cross-training colleagues to cover for them. recruiting replacements.

Show thread

aliengasmask 12h ago

@carnage4life yes! All skills and other guff need reproducible tests to verify the black box still does what it did when you wrote them.

Ive seen none at all, anywhere.