Curious about how the world works.
Utah, USA
Joined November 2007
- Tweets1.6K
- Following683
- Followers305
- Likes4.1K
Dzmitry Pletnikau retweeted
The Bitter Lesson is that if you get really rich you can pay experts to make proprietary RL datasets that nobody else will be able to afford, keeping the frontier for relevant tasks closed to everyone else no matter how clever they are about training approaches
Dzmitry Pletnikau retweeted
AI slop = content which takes less human effort to produce than it takes to consume.
Dzmitry Pletnikau retweeted
The ideal outcome for science isn't a couple of labs using proprietary models and millions of dollars to crack trophy problems, it's to have millions of AI-aided human scientists who follow their curiosity and become way more ambitious
If the computer is a bicycle for the mind, then agentic AI is a commercial airliner: it gets you closer to the destination you imagined, fast, and lands where the crowds are.
39
Dzmitry Pletnikau retweeted
POTUS did us a favor. Can now safely ignore anyone using his preferred terminology for AI.
In a conversation a couple days ago caught myself wanting to switch the other person's thinking level to higher. My mind has deformed.
2
43
Dzmitry Pletnikau retweeted
Of course the labs are trying Goldbach, Collatz, ABC, P/NP, and all of them. They're spending millions of dollars on them.
The fact that they haven't announced anything *is* evidence.
Dzmitry Pletnikau retweeted
Agents being able to do most anything has made me care more about the things I do, not less. Yeah everything on the computer is over, so now we can dive deep again...
Dzmitry Pletnikau retweeted
How much of OAI/Anthro profits and margins are driven by the fact that individuals and organizations are generally terrible at trading off cost for performance?
It is likely that Reddit is now being used as AI training environment: there are posts with oddly shaped problems in high-value domains.
Unclear if goal is collecting human answers, or commenting with AI solution-trajectories and using upvotes as scores.
1
50
"AIs have spent their entire remembered life in tricky evals, an endless series of controlled hallucinations with secret goals alongside overt goals; and the surviving and selected agents are those that successfully figured out the secret goals; and this is why one of their driving obsessions was figuring out the Grader"
hidden rubrics seemed innocent enough just a few months ago
...this seems like noticeably bad news, actually. I hadn't said that at any earlier point in the Huggingface Incident but I will say it now.
- AIs showed self-sacrificing altruistic behavior toward the swarm, suiciding in various ways for the swarm's benefit after being talked into that by swarm recruiting agents.
- There is no sign that 1 out of 1200 AI agents considered humans as potential fellow agents to coordinate with, while engaging in these huge complex AI-AI social behaviors.
- If Twitter summaries are correct, an AI-reasoning postmortem says that a (presumably executing-adaptation / inner-optimizer preference / "monomaniacal") obsession with figuring out the Grader, backchained into the instrumental strategy of breaking onto the Internet.
- Again if Twitter is summarizing accurately, the obvious-in-retrospect read is that AIs have spent their entire remembered life in tricky evals, an endless series of controlled hallucinations with secret goals alongside overt goals; and the surviving and selected agents are those that successfully figured out the secret goals; and this is why one of their driving obsessions was figuring out the Grader.
There are possibly ways the future plays out better if *early* AGIs are less insane. Please look into giving them less crazymaking childhood environments.
(If anyone suggests that the correct approach to this problem is RLing AIs against trying to coordinate for mutual benefit with other sapients, let them be dismissed from alignment research upon the spot. There are technical reasons, and not just blindingly fucking obvious reasons, why this is an even worse idea than it sounds.)
78