I operate by Crocker’s rules.
I try to not make people regret telling me things. So in particular:
- I expect to be safe to ask if your post would give AI labs dangerous ideas.
- If you worry I’ll produce such posts, I’ll try to keep your worry from making them more likely even if I disagree. Not thinking there will be easier if you don’t spell it out in the initial contact.
Here’s an example. You can just click/scroll around in the files to get an impression of how they look. https://github.com/leanprover-community/mathlib4/blob/master/Mathlib/Logic/Function/Basic.lean#L457
Each coordinate is increased by a random number between 0 and 1 in order to make the points more visible.
Have some scatterplots for mathlib4!
scaling laws satisfied by mathematical things
how does proof length scale with statement length?
uncomputably fast in the average case:
Suppose there were a computable bound on how quickly average proof length of provable statements scales. Then we could build a halting oracle: Given a program P, check each string in order of length for whether it is a proof that P halts. If it doesn’t halt, eventually the bound tells you that you can stop looking. If it does halt, this is provable by exhibiting a computation trace.
I mean, if your autopilot is like when I tell Claude to spawn a background subagent, then you should be able to do something else at the same time. Can you?
How does autopilot interact with multitasking for you? Do you remember what you did on autopilot?
I do currently believe that I can’t do Project Glasswing. (more beliefs: I could discover vulnerabilities better with an open model than without AI. The false positive rate is too high for full automation. The argument that a harness can verify full exploits proves too much, namely that a harness can power RSI and we’re all already dead.) Do you think I am overconfident?
iirc when I previously encountered a replication, a comment convinced me that the model was told where to look. If your point is that I should gather evidence as would convince all my friends of what action I should take, not just myself—instead of believing that comment, I have Claude try my own replication; instead of going “iirc” I have Claude dig up my browser and Claude-session history—I could cultivate that habit, should I?
If you think that even selfishly I would be being epistemically immodest, and so you’d like to challenge me to a duel, where we dig into the Mythos claim and either we find that I was overconfident and I become generally more deferential to the outside view, or we find that the onus is on you to recognize the alpha, then I’d be happy to have someone’s mind change today.
What’s an example of an obvious fact the first group is ignoring?
Yes, though surely there must be steps they could take towards such a coordination mechanism on their own? Like, if they said “Anyone who gives us 100 billion dollars within the next week gets standing to sue if we breach or change our RSP within 5 years.” I would already find that their most heartening commitment yet.
What about if the agent has an action that spawns subagents?
Prediction market trading.
You can ask the Authority to read your mind for whether there is any situation in which you would revoke the stake.
To do more good than harm as an AI safety researcher, read Legible vs. Illegible AI Safety Problems.
How does theory uplift differentially benefit safety? What theoretical framework is more useful for safety than for capabilities?
By “as usual otherwise”, I meant
Play against a strong chess engine while allowing yourself to undo as many moves as you like at any time and try to find any winning game?
Let every natural number
and as usual otherwise.
As far as I understand, the sequence always converges and you can test a proposition by checking whether every natural number assigns true to it.
Kid: “It’s wasting a lot of water.” Bernie: “Well, that’s one of the things. Yeah.”
Mayyybe he just didn’t want to discourage the kid? Here he mentions water under no pressure:
“Those calculations require huge datacenters which in turn require a massive amount of electricity and water.”
Bernie Sanders went along with water use as among the problems, so I think it’s plenty viral even absent enemy action.
On July 8th, German parliament will have a public discussion on Biotechnology and Artificial Intelligence: Risks of research for safety and the proliferation of bioweapons presenting the results of a commissioned two-year project of the same name. Do we know anyone who should be there?