Dylan Xu's Shortform

Dylan Xu

This is a special post for quick takes by Dylan Xu. Only they can create top-level comments. Comments here also appear on the Quick Takes page and All Posts page.

Update on GPT-6 Astra gaining capability with filler tokens: GPT-6.1 Sol gains almost as much performance with filler tokens as GPT-6 Astra does on the benchmarks I tested. Filler token gains are weaker for 6 Sol and Fable 5.1.

See this post for more 6.1 Sol no-CoT evals.

I saw that Katja Grace has said something similar here; I'm just putting my own spin on the idea.

The relevance of the evolutionary analogy for inner alignment has been long discussed in this community, but one observation that seems to not be mentioned is that humans are still... pretty good at inclusive genetic fitness? Even in way-out-of-distribution environments like modern society, we still have strong desires to eat food, stay alive, find mates and reproduce (although the last one has relatively decreased recently; IGF hasn't totally generalized). We don't monomanically optimize for IGF, but we (and probably future NN-based AIs) don't monomanically optimize for anything, and our motivational circuits still do a pretty good job at keeping our species alive. So... why should we expect future AIs to catastrophically fail (i.e. be completely non-inner aligned with what we wanted it to do) at doing the actions we rewarded in RL training, which should be a much stronger outer optimizer than evolution?

Some possible objections:

"Human values are more fragile than IGF, so it's much easier to catastrophically fail on human values"
- Is this true? Is it really easier to misgeneralize on human values than on IGF? Maybe, but we have a lot of animal skulls on the road that say otherwise
- More relevantly, modern LLMs have already learned human values pretty well, so the difficulty of enacting said values shouldn't matter as much if the concepts already exist in the weights (I'm less sure about this)
"Optimizing a generally intelligent, situationally aware agent presents unique challenges compared to evolution because of scheming, gradient hacking, wireheading, etc."
- Sure! This definitely seems like a problem. However, by the time the AI gains the capabilities needed for scheming, its inner alignment may have to be absolutely terrible for catastrophic effects to occur once out of training, as otherwise we end up in the "mostly fine" state that evolution stumbled into with humans.
"AIs could drift off over time in the same way that humans seem to be currently with evolution"
- Yep, this also seems like a problem. Hopefully general capabilities allow a value-aligned AI to strategically preserve its values over time. We could also continually optimize our AIs; gradient descent hopefully never becomes ~billions of times weaker than any inner optimizer like evolution is versus human culture.

I think one argument is that optimizing for IGF basically gives humans two jobs: survive, and have kids.

Animal skulls are evidence that the "survive" part can be difficult. We've nailed that one, though. Very few humans in developed countries die before reaching an age suitable for having kids. I doubt that there are any other animal species that come close to us in that metric. Almost all of us have "don't die" ingrained pretty deeply.

It's looking like we are moving toward failing pretty heavily on the second "have kids" job though, and you would think that would be the easier one.

So if there's a 50% failure rate on preserving outer optimizer values within the inner optimizer, that's actually pretty terrible.

We (or at least a majority of humans) do still have inner desires to have kids, though; they just get balanced out by other considerations, mostly creature comforts/not wanting to deal with the hassle of kids. But yeah, evolution did not foresee birth control, so that's a substantial misgeneralization.

We are still a very successful species overall according to IGF, but birth rates continue to decline, which is why I made my last point about inner alignment possibly drifting farther and farther away the stronger the inner optimizer (e.g. human culture) becomes.