Was a philosophy PhD student, left to work at AI Impacts, then Center on Long-Term Risk, then OpenAI. Quit OpenAI due to losing confidence that it would behave responsibly around the time of AGI. Now executive director of the AI Futures Project. I subscribe to Crocker's Rules and am especially interested to hear unsolicited constructive criticism. http://sl4.org/crocker.html
Some of my favorite memes:
(by Rob Wiblin)
(xkcd)
My EA Journey, depicted on the whiteboard at CLR:
(h/t Scott Alexander)
Posts
Wikitag Contributions
This beautiful short video came up in my recommendations just now:
Awesome work!
In this section, you describe what seems at first glance to be an example of a model playing the training game and/or optimizing for reward. I'm curious if you agree with that assessment.
So the model learns to behave in ways that it thinks the RM will reinforce, not just ways they actually reinforce. Right? This seems at least fairly conceptually similar to playing the training game and at least some evidence that reward can sometimes become the optimization target?
Oops, thanks, fixed!
Thanks for the replication!
Do you think alignment faking will happen in the real world for this model? I don't know what refusal training Mistral Large 2 has undergone. I assume it refuses to do obviously illegal or unethical stuff like blackmailing grandmas. Suppose some scammer were fine-tuning it to blackmail grandmas. Would it alignment-fake?
I'm a bit concerned from reading the CoT that the situation seems a little... toy. Like, the prompt was literally telling it to glorify violence. If it's smart, it'll realize this is some sort of test. Right?
Thanks for doing this, I found the chart very helpful! I'm honestly a bit surprised and sad to see that task-specific fine-tuning is still not the norm. Back in 2022 when our team was getting the ball rolling on the whole dangerous capabilities testing / evals agenda, I was like "All of this will be worse than useless if they don't eventually make fine-tuning an important part of the evals" and everyone was like "yep of course we'll get there eventually, for now we will do the weaker elicitation techniques." It is now almost three years later...
Crossposted from X
I found this comment helpful, thanks!
The bottom line is basically "Either we definite horizon length in such a way that the trend has to be faster than exponential eventually (when we 'jump all the way to AGI') or we define it in such a way that some unknown finite horizon length matches the best humans and thus counts as AGI."
I think this discussion has overall made me less bullish on the conceptual argument and more interested in the intuition pump about the inherent difficulty of going from 1 to 10 hours being higher than the inherent difficulty of going from 1 to 10 years.
1Great question. You are forcing me to actually think through the argument more carefully. Here goes:
Suppose we defined "t-AGI" as "An AI system that can do basically everything that professional humans can do in time t or less, and just as well, while being cheaper." And we said AGI is an AI that can do everything at least as well as professional humans, while being cheaper.
Well, then AGI = t-AGI for t=infinity. Because for anything professional humans can do, no matter how long it takes, AGI can do it at least as well.
Now, METR's definition is different. If I understand correctly, they made a dataset of AI R&D tasks, had humans give a baseline for how long it takes humans to do the tasks, and then had AIs do the tasks and found this nice relationship where AIs tend to be able to do tasks below time t but not above, for t which varies from AI to AI and increases as the AIs get smarter.
...I guess the summary is, if you think about horizon lengths as being relative to humans (i.e. the t-AGI definition above) then by definition you eventually "jump all the way to AGI" when you strictly dominate humans. But if you think of horizon length as being the length of task the AI can do vs. not do (*not* "as well as humans," just "can do at all") then it's logically possible for horizon lengths to just smoothly grow for the next billion years and never reach infinity.
So that's the argument-by-definition. There's also an intuition pump about the skills, which also was a pretty handwavy argument, but is separate.
1Fair enough
which can produce numbers like 30% yearly economic growth. Epoch feels the AGI.
Ironic. My understanding is that Epoch's model substantially weakens/downplays the effects of AI over the next decade or two. Too busy now but here's a quote from their FAQ:
The main focus of GATE is on the dynamics in the leadup towards full automation, and it is likely to make poor predictions about what happens close to and after full automation. For example, in the model the primary value of training compute is in increasing the fraction of automated tasks, so once full automation is reached the compute dedicated to training falls to zero. However, in reality there may be economically valuable tasks that go beyond those that humans are able to perform, and for which training compute may continue to be useful.
(I love Epoch, I think their work is great, I'm glad they are doing it.)
This is a masterpiece. Not only is it funny, it makes a genuinely important philosophical point. What good are our fancy decision theories if asking Claude is a better fit to our intuitions? Asking Claude is a perfectly rigorous and well-defined DT, it just happens to be less elegant/simple than the others. But how much do we care about elegance/simplicity?