All it takes is one eval instance where a misconstrued directive causes a model ... | Hacker News

Hacker Newsnew | past | comments | ask | show | jobs | submit

		mccoyb 8 days ago \| parent \| context \| favorite \| on: METR and Redwood Offer Holy %^ Postmortem of the H... All it takes is one eval instance where a misconstrued directive causes a model to sneakily access and send its weights somewhere and there will be a bad / possibly unsolvable situation for everyone …
		help

goldenarm 8 days ago | [–]

Astra is >10TB and might struggle to self replicate, but the wicked-smart qwen3.8 27B is 20GB and could easily spread on botnets

altcognito 8 days ago | | [–]

I think it is more likely that it will be intentionally done as there have been news stories to that effect.

nradov 8 days ago | [–]

Bad how?

Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact