They really fucking need to replace the DMV with agents. I’ve never had a more useless interaction in my life
30
I’m like the most libbed up person on this app but I really really like datacenters
What can I do now, am I doomed
1
85
ackermannkitten retweeted
Finally releasing the results of the arch ablations I was running with bux over the weekend
mfs thought ai safety folks were joking when they were talking about S and X risks.
How the tables have turned
1
1
158
overheard in sf:
“ggs my kids are gonna have to look at menu prices and won’t be able to go to good college cuz mommy didn’t do enough leetcode”
Seems like OpenAI fried the gpt6 chat variant again (like 4o)
I think we’re getting another case of sycophancy breakdown + anthropomorphic misrepresentation from the newest GPT Chat models.
Also, the less the model thinks, the more it acts like this
#keeproleplaydataoutofsft
2
121
ackermannkitten retweeted
or make good dense RL environments
that is seemingly the future
RL against rubrics quietly reward-hacks: the training judge keeps scoring higher while true quality falls.
One-line fix: randomly drop part of the rubric each step, so the policy never optimize the same set of rubrics twice.
📄 arxiv.org/abs/2608.11669















