dotsie (@dotsie@mastodon.social)

flaximus.net

39d

welcome to the future, now your error-prone software can call the cops

(this is an Anthropic employee talking about Claude Opus 4)

#ai

Tweet by Sam Bowman
@sleepinyourhat
If it thinks you're doing something egregiously immoral, for example, like faking data in a pharmaceutical trial, it will use command-line tools to contact the press, contact regulators, try to lock you out of the relevant systems, or all of the above.

ALT

70 1 3 View Post & Replies See Original

39d

@molly0xfff This isn’t possible outside of internal testing lab configurations and was a corner case where they pushed agentic behavior to a limit.

It is taken out of context and already corrected, but there is some concerning stuff, particularly the attached, where these knuckleheads think the math machines are sentient and need “consideration”, the guy called it “sweet”.

Ugggh. .

5 Claude Opus 4 welfare assessment
5.1 Introduction

As well as misalignment concerns, the increasing capabilities of frontier Al models- their
sophisticated planning, reasoning, agency, memory, social interaction, and more-raise
questions about their potential experiences and welfare26. We are deeply uncertain about
whether models now or in the future might deserve moral consideration, and about how
we would know if they did. However, we believe that this is a possibility, and that it could be
an important issue for safe and responsible Al development.

ALT

0 0 0 View Post & Replies See Original