FleetingBits
banner
fleetingbits.bsky.social
FleetingBits
@fleetingbits.bsky.social
Are base models the dreams of an LLM?
October 19, 2025 at 1:38 AM
July 22, 2025 at 4:45 PM
June 22, 2025 at 2:20 AM
inspiration
December 4, 2024 at 2:09 AM
December 4, 2024 at 2:06 AM
December 3, 2024 at 5:51 AM
or, the data labelers don't check the 20 line code fragment that the model spit out

hmmm
December 2, 2024 at 4:47 AM
And, then the finding ends up being something like "our data labelers don't check the references" so we get bad labels : X

uhh, ok
December 2, 2024 at 4:44 AM
this paper is referenced, which should have examples of reward hacking and the authors are high quality authors (Ethan Perez!)
December 2, 2024 at 4:44 AM
I think in your example - people would demand their money back right now if the agent failed.

I'm not sure how many agents will be described as agents in the future - or just as services - in which case, all of this is moot.
December 2, 2024 at 12:22 AM
December 1, 2024 at 5:52 PM
December 1, 2024 at 4:31 PM
November 30, 2024 at 11:01 PM
November 30, 2024 at 5:24 PM
November 15, 2024 at 12:48 PM
November 14, 2024 at 4:19 PM
November 14, 2024 at 3:32 PM
I'm starting to really wonder if the issue at OpenAI with safety is that the safety advocates don't understand what it's like to work in a "control function" at a company.

It's pretty brutal work.
November 14, 2024 at 9:36 AM
November 14, 2024 at 9:26 AM
November 13, 2024 at 7:20 PM