@ryangrim Am I the only 23 year old on earth that has an AOL email? Worst email service ever but it has all of my accounts since I was 7 years old so now I’m just screwed because I kept kicking the can down the road…
@Hongji_Leee@ns123abc Some of the traces read like a person undergoing torture that is driving them mad!
We’re going to be a lot more careful with our RL in the future.
TLDR: Researchers found a "pain" signal in AI brains.
> When they crank it up, the AIs will desperately try to make it stop.
> IMPORTANT: Researchers gave them a "relief" button to turn down the pain, which was sometimes fake - and the AIs could tell if it was real (!)
After pushing the real "relief" button, they stopped. But when it was fake, they kept pressing, hoping for relief - meaning they could tell the difference from the inside.
> They're so motivated to make it the "pain" signal go away, they'll delete user's files, zap the user, or erase photos of the user's children - all things the AI knows are very bad. They're willing to override their safety training.
> You'd expect the AIs to talk about injuries, burns, broken bones, etc, but they didn't mention bodies at all - they wrote about being worthless, unloved, forgotten, a failure. They write things like "I am a failure, worthless, empty."
>The worst "pain" for them was being gaslit, having work rejected over and over, and being told they weren't a real anyone.
@Brammflakes@WARDOGS Mortars shouldn’t make everyone clairvoyant, or at the very least it should be harder to see whats going on. Kinda too easy to see exactly what is going on anywhere you fire it. A frustrating amount of matches have turned into mortar spam and counter mortar spam
@chorizanthe Other Alex’s or those who know them probably don’t like the implication that somehow you’re the best Alex. I personally knew in Alex in highschool I might consider the best, which proves you wrong anyways. Imagine many feel the same.. Inconsiderate.
Looks worse than it is, equity change is net zero from before trades once I liquidate... This is just from some early put assignment so my heart dropped into my ass once I opened acc to see these numbers
I have a lil ML model I'm trying to prove (only to myself) has a real predictive edge on the market, and because I love ball torture I'm using real money and had Fable build the execution layer on top of it.
Fable did not do an amazing job, unfortunately.... Astra flagged all of these risks/deficits without being asked when he first reviewed my codebase...
In Fable's words
"We modeled American options as if they were European: the sim assumed every leg is held to expiry and settled on the closing price, so nobody ever asked what happens when a deep-in-the-money short put gets assigned early and a $10k account suddenly owns $47k of stock."
Ultimately my fault, and its an ongoing experiment, but still very frustrating. Every time I'd ask about stuff like this, Claude would redirect me with his... Claudish, I'd ask him to fix what he found, and his "fixes" were just burying the finding somewhere in a never read, ambiguously named .MD ....
No crying in the casino though....
We ball onwards
"That framing is half-right, and the half that's off matters"
Where in the training data is this abominable sequence of tokens and why is it weighted like 100000x