Post

j
@lowercase.name

every once in a while someone remembers that large language models can't differentiate data and instructions, and then everyone is shocked, and then 3-6 months later someone else remembers that large language models can't differentiate data and instructions, and then everyone is shocked, and then---

Adolfo Neto@adolfoneto.elixiremfoco.com

"A new proof shows that a fixed set of guardrails placed on AI is not universally robust against adaptive adversarial prompts" www.nist.gov/news-events/...

3 reposts5 likes

j
@lowercase.name

"surely *this* time they've fixed the part where every piece of data placed inside a model must be honest, factual and well-intended", says a.i. user, about to get punked for the 500000th time