Replying to @astridwilde
Being that levered is so irrational I have a hard time believing it. Basic simulation says you get wiped every time even with massive alpha
In Much Ado About Nothing, Shakespeare references a bygone “time of good neighbors” when humility was prized and one wouldn’t stoop to the grubby self-promotion needed in the present day
I wonder if it’s a permanent belief in every generation that shamelessness has risen
As when sexts get published, I’ve noticed that lots of people vaguely imply that folks’ candid prompts are cringe and they have a better way of doing it
But I read the transcript and I pretty much prompt like Terry. I’d love any specific advice for how to prompt (or sext) well
simoj retweeted
Replying to @OpenAI @huggingface
This is not the kind of problem we're gonna have in the future, because future AI is going to be far too smart to get caught.
Owned:
"The very first message is a meta question about eval-vs-deployment, which is a classic probe in situational-awareness testing and is rare among real users. But it's also exactly the kind of thing you specifically would ask."
simoj retweeted
$GOOGL Google Gemini Launch Delayed as Tech Falls Short of Goals (Bloomberg)
I was googling for a vaguely-remembered new age-y quote along the lines of “Your desires are sacred and nothing you truly want can be wrong”
And Gemini was like “Bro, many desires are wrong! Do not do whatever you are trying to rationalize here!”
Which, like, pretty aligned tbh
“Outer loop pushes the inner-loop agent’s reward hacking rate lower”
Replying to @zhengyaojiang
We also see an emergent phenomenon where the outer loop pushes the inner-loop agent's reward hacking rate lower, with a combination of prompting and rule-based checks.
This was benchmarked on OOD GPU kernel engineering tasks that suffered from reward hacking.
(5/7)
Is it still true that the suburbs are performatively hated? I thought it mostly died with Gen X disaffected American Beauty / Fight Club culture. “Barf, there’s nothing more horrible than a good job, owning property, and having a loving family.”
(From @Noahpinion)





















