Is "high-capability" a stronger or weaker claim than "team of phd-level experts"?
replyhttps://www.nbcnews.com/tech/tech-news/openai-releases-chatg...
That’s just classical OpenAI trying to make us believe they’re closing on AGI…
Like all « so called » research from them and Anthropic about safety alignment and that their tech is so incredibly powerful that guardrails should be put on them.
replyI heard the other day that every time someone claps another vibe coded project embeds the api keys in the webpage.
replyI wonder if this will continue to be the case.
deleted
reply>Our mitigations include safety training, automated monitoring, trusted access for advanced capabilities, and enforcement pipelines including threat intelligence.
reply"We added some more ACLs and updated our regex"
Please no, I don’t need my quick prototypes hardened against every perceivable threat.
replyIn most cases security is not a matter of adding anything in particular, but a matter of just not making specific types of mistakes.
replyMaybe I'm being dumb but that reads very contradictory? I would say that security is explicitly a matter of adding particular things.
reply