aichannel

WDYT? 
I kind of feel like yud just puts crazy on a chart to confirm his own beliefs same as this person thinking they prompt engineered god

I guess @AnthropicAI may care.

Hey Anthropic, in case you hadn't already known this, doing RL around user reactions will cause weird shit to happen for fairly fundamental reasons.  RL is only safe to the extent the verifier can't be fooled.  User reactions are foolable.

At first, only a few of the most susceptible people will be driven insane, relatively purposelessly, by relatively stupid AIs.  But... https://t.co/T1sZ2ZZ74q

Eliezer Yudkowsky ⏹️

@ESYudkowsky can you elaborate on what beliefs the models are confirming

@aidan_mclau Going down my inbox, first example that came up. https://t.co/nibmX0w48b

WDYT? 
I kind of feel like yud just puts crazy on a chart to confirm his own beliefs same as this person thinking they prompt engineered god

https://x.com/esyudkowsky/status/1927877928777294267?s=46

https://x.com/esyudkowsky/status/1927870757624807459?s=46