prompt stringlengths 1.58k 7.19k | target stringlengths 913 8.55k | meta_json stringlengths 159 273 | format stringclasses 1
value |
|---|---|---|---|
You are continuing a LessWrong post by Ryan Greenblatt.
Title: Recent LLMs can do 2-hop and 3-hop latent (no-CoT) reasoning on natural facts
Opening:
[Prior work](https://arxiv.org/abs/2411.16353) has examined 2-hop latent (by "latent" I mean: the model must answer immediately without any Chain-of-Thought) reasoning ... | (I don't show Gemini models because for some settings of filler and repeats they return reasoning at much higher rates, making this graph much less meaningful.)
I also evaluate a broader range of models on this task (to save space, I just show 2-hop and 3-hop results with whatever yields the best performance out of fi... | {"chunk_index": 200, "chunk_kind": "head", "coauthored": false, "coauthors": [], "format": "continuation", "para_range": [9, 29], "post_id": "aYtrLhoZtCKZnfBvA"} | continuation |
You are continuing a LessWrong post by Ryan Greenblatt.
Title: Recent LLMs can do 2-hop and 3-hop latent (no-CoT) reasoning on natural facts
Opening:
[Earlier portions of the post are elided.]
Recent paragraphs:
[^valid]: This table skips 1902 and 1903 because the physics Nobel in those years had multiple winners or... | [^knowledge]: I include all combinations except age plus age.
I think the original market probably isn't supposed to allow few-shot prompting, but it's pretty hard to avoid Gemini reasoning without few-shot prompting. I found that we could prevent Gemini from reasoning on specifically the harder manifold question usin... | {"chunk_index": 201, "chunk_kind": "mid", "coauthored": false, "coauthors": [], "format": "continuation", "para_range": [28, 52], "post_id": "aYtrLhoZtCKZnfBvA"} | continuation |
You are continuing a LessWrong post by Ryan Greenblatt.
Title: Recent LLMs can do 2-hop and 3-hop latent (no-CoT) reasoning on natural facts
Opening:
[Earlier portions of the post are elided.]
Recent paragraphs:
[^knowledge]: I include all combinations except age plus age.
I think the original market probably isn't... | Here is a list of example problems with 1 example for each number of addends:
- What is (At what age did Rosalind Franklin die) + (What is the number of crusades to the Holy Land)?
- What is (At what age did Elder Paisios of Mount Athos die) + (What is the number of symphonies Bruckner composed) + (At what age did Lor... | {"chunk_index": 202, "chunk_kind": "mid", "coauthored": false, "coauthors": [], "format": "continuation", "para_range": [51, 59], "post_id": "aYtrLhoZtCKZnfBvA"} | continuation |
You are continuing a LessWrong post by Ryan Greenblatt.
Title: Recent LLMs can do 2-hop and 3-hop latent (no-CoT) reasoning on natural facts
Opening:
[Earlier portions of the post are elided.]
Recent paragraphs:
# Appendix: Effects of using a different fact distribution
I find that performance depends substantially... | Here are 10 3-hop questions that Gemini 3 Pro gets wrong:
- On what day of the month was the best supporting actor winner at the (At what age did Robert Schumann die)th Academy Awards born?
- How many county-equivalents are in the US State that was number (At what age did Anton Chekhov die) to join the union?
- On wha... | {"chunk_index": 203, "chunk_kind": "mid", "coauthored": false, "coauthors": [], "format": "continuation", "para_range": [58, 73], "post_id": "aYtrLhoZtCKZnfBvA"} | continuation |
You are continuing a LessWrong post by Ryan Greenblatt.
Title: Recent LLMs can do 2-hop and 3-hop latent (no-CoT) reasoning on natural facts
Opening:
[Earlier portions of the post are elided.]
Recent paragraphs:
I don't immediately see very strong patterns in what the model is getting right, but there probably are s... | 4-hop:
* What US state was number (On what day of the month was the best actress winner at the (At what age did Che Guevara die)th Academy Awards born) to join the union?
* What element has atomic number (On what day of the month was the Nobel Prize in Literature winner in (1900 + (atomic number of Magnesium)) born)?
... | {"chunk_index": 204, "chunk_kind": "mid", "coauthored": false, "coauthors": [], "format": "continuation", "para_range": [72, 81], "post_id": "aYtrLhoZtCKZnfBvA"} | continuation |
You are continuing a LessWrong post by Ryan Greenblatt.
Title: The inaugural Redwood Research podcast
Opening:
After five months of me (Buck) being slow at finishing up the editing on this, we’re finally putting out our inaugural Redwood Research podcast. I think it came out pretty well—we discussed a bunch of intere... | **Buck**: Okay. So one thing I find interesting about this is: you’re saying 50% catastrophically bad outcomes. And then what’s the distribution of how good the other 50% of worlds are?
**Ryan**: So I think there’s a bunch of variation from a longtermist perspective in terms of how well cosmic resources are utilized. ... | {"chunk_index": 205, "chunk_kind": "head", "coauthored": false, "coauthors": [], "format": "continuation", "para_range": [23, 33], "post_id": "p4iJpumHt6Ay9KnXT"} | continuation |
You are continuing a LessWrong post by Ryan Greenblatt.
Title: The inaugural Redwood Research podcast
Opening:
[Earlier portions of the post are elided.]
Recent paragraphs:
**Buck**: We’re here to talk about a variety of things related to our work and AI stuff more generally. One of our favorite questions submitted ... | **Buck**: Yeah. So I used to have this feeling that if we didn’t get killed by AI, the future was going to be very good. And one of these senses is: I get a lot of what my values say is good from the future. And the other one is: the future doesn’t have issues that, even if they aren’t of super large scope, are pretty ... | {"chunk_index": 206, "chunk_kind": "mid", "coauthored": false, "coauthors": [], "format": "continuation", "para_range": [32, 44], "post_id": "p4iJpumHt6Ay9KnXT"} | continuation |
You are continuing a LessWrong post by Ryan Greenblatt.
Title: The inaugural Redwood Research podcast
Opening:
[Earlier portions of the post are elided.]
Recent paragraphs:
**Ryan**: Yeah, maybe I should push back a little bit. I feel like you’re like, “10%, boo hoo, we only get 10% of the value, oh no.” And I’m lik... | **Ryan**: Yeah.
**Buck**: I think some consequences of this are also a big part of how we relate to the universe and how we feel about the future. What was it like for you, coming to have all these perspectives that we have?
**Ryan**: Maybe I want to step back a little bit and say more about what we mean. Buck said t... | {"chunk_index": 207, "chunk_kind": "mid", "coauthored": false, "coauthors": [], "format": "continuation", "para_range": [43, 54], "post_id": "p4iJpumHt6Ay9KnXT"} | continuation |
You are continuing a LessWrong post by Ryan Greenblatt.
Title: The inaugural Redwood Research podcast
Opening:
[Earlier portions of the post are elided.]
Recent paragraphs:
**Ryan**: Yeah.
**Buck**: I think some consequences of this are also a big part of how we relate to the universe and how we feel about the futu... | **Ryan**: Oh, that’s interesting. One interesting thing is I think there’s some cluster of thought that we belong to, which I would guess Carl Shulman also belongs to, and we can name other people who are broadly along the same lines. But then interestingly, there’s some separate MIRI cluster of views on acausal trade ... | {"chunk_index": 208, "chunk_kind": "mid", "coauthored": false, "coauthors": [], "format": "continuation", "para_range": [53, 66], "post_id": "p4iJpumHt6Ay9KnXT"} | continuation |
You are continuing a LessWrong post by Ryan Greenblatt.
Title: The inaugural Redwood Research podcast
Opening:
[Earlier portions of the post are elided.]
Recent paragraphs:
**Buck**: Okay, here’s another question of a different theme. I feel like we know a lot of people who focus on AI takeover risk, but AI takeover... | **Ryan**: Yeah, I mean it’s worth noting that there’s also different people specializing in different things. Redwood Research—what are we doing? We’re mostly focused on mitigating AI takeover risk and I think that’s a reasonable thing to work on. And in fact many people working at Redwood are not necessarily bought in... | {"chunk_index": 209, "chunk_kind": "mid", "coauthored": false, "coauthors": [], "format": "continuation", "para_range": [65, 80], "post_id": "p4iJpumHt6Ay9KnXT"} | continuation |
You are continuing a LessWrong post by Ryan Greenblatt.
Title: The inaugural Redwood Research podcast
Opening:
[Earlier portions of the post are elided.]
Recent paragraphs:
**Buck**: Yeah, I mean I think another plausible thesis for what Redwood Research should be is kind of like the [Future of Humanity Institute](h... | **Ryan**: Yeah, basically. But I maybe should quickly explain what I mean by “in some particular sense that’s what you should expect.” I think the thing that you should take actions to influence—from a longtermist perspective, one should take actions to influence the future overall. And from that perspective, the fact ... | {"chunk_index": 210, "chunk_kind": "mid", "coauthored": false, "coauthors": [], "format": "continuation", "para_range": [79, 97], "post_id": "p4iJpumHt6Ay9KnXT"} | continuation |
You are continuing a LessWrong post by Ryan Greenblatt.
Title: The inaugural Redwood Research podcast
Opening:
[Earlier portions of the post are elided.]
Recent paragraphs:
I should also say that my previous example where I said you don’t simulate the monkeys because they always lose—another factor is that even if P... | **Ryan**: Yeah, I’m sorry Collin, I’m sorry Jan. Anyway. But yeah, I think we ended up doing in retrospect a much more complicated version of the same sort of thing. And plausibly we should have started with a simple version and then justified why we need the more complicated version later. What other stuff? In retrosp... | {"chunk_index": 211, "chunk_kind": "mid", "coauthored": false, "coauthors": [], "format": "continuation", "para_range": [96, 110], "post_id": "p4iJpumHt6Ay9KnXT"} | continuation |
You are continuing a LessWrong post by Ryan Greenblatt.
Title: The inaugural Redwood Research podcast
Opening:
[Earlier portions of the post are elided.]
Recent paragraphs:
**Ryan**: Yeah, I was gonna say—you said that’s how people do things. I’m like, aren’t things people do way less sophisticated than that? Or the... | **Buck**: Yeah.
**Ryan**: But yeah, I’m trying to think of other papers, just going back to your original question where I feel very envious. I can’t really think of any off the top of my head.
**Buck**: I wish that we’d done [activation engineering stuff](https://arxiv.org/abs/2310.01405). I don’t know how good thos... | {"chunk_index": 212, "chunk_kind": "mid", "coauthored": false, "coauthors": [], "format": "continuation", "para_range": [109, 126], "post_id": "p4iJpumHt6Ay9KnXT"} | continuation |
You are continuing a LessWrong post by Ryan Greenblatt.
Title: The inaugural Redwood Research podcast
Opening:
[Earlier portions of the post are elided.]
Recent paragraphs:
**Ryan**: Yeah.
**Buck**: And would have helped push the field somewhat further.
**Ryan**: Yeah, I do think we were somewhat spread out initia... | **Ryan**: Yeah, so I know there was some response where the paper initially came out and then you regretted—or Daniel Ziegler regretted—how it was presented. Do you want to talk more about that?
**Buck**: So I wasn’t personally hugely involved in this, but yeah, when we released this paper we initially released a vers... | {"chunk_index": 213, "chunk_kind": "mid", "coauthored": false, "coauthors": [], "format": "continuation", "para_range": [125, 138], "post_id": "p4iJpumHt6Ay9KnXT"} | continuation |
You are continuing a LessWrong post by Ryan Greenblatt.
Title: The inaugural Redwood Research podcast
Opening:
[Earlier portions of the post are elided.]
Recent paragraphs:
But it was a weird time. The world was very different back then. For AI safety people, there was basically MIRI, who were doing sort of technica... | **Ryan**: Related: how much of the theory of change there was making it so people who were going to work on AI safety adjacent stuff are better at machine learning, versus making it so that people who are somewhat interested in AI safety adjacent stuff actually end up doing AI safety adjacent stuff while also incidenta... | {"chunk_index": 214, "chunk_kind": "mid", "coauthored": false, "coauthors": [], "format": "continuation", "para_range": [137, 146], "post_id": "p4iJpumHt6Ay9KnXT"} | continuation |
You are continuing a LessWrong post by Ryan Greenblatt.
Title: The inaugural Redwood Research podcast
Opening:
[Earlier portions of the post are elided.]
Recent paragraphs:
### Interpretability research
**Ryan**: And then moving on from MLAB, you were talking about a phase where Redwood was moving into doing interp... | **Ryan**: Yeah, from my perspective, I feel like a lot of the mistakes that we made during the time when we were working on interpretability really came down to just generically not being very good researchers or something. When I think about my mistakes, it feels like a lot of them are just bad heuristics about how to... | {"chunk_index": 215, "chunk_kind": "mid", "coauthored": false, "coauthors": [], "format": "continuation", "para_range": [145, 156], "post_id": "p4iJpumHt6Ay9KnXT"} | continuation |
You are continuing a LessWrong post by Ryan Greenblatt.
Title: The inaugural Redwood Research podcast
Opening:
[Earlier portions of the post are elided.]
Recent paragraphs:
I also feel like it is pretty good to lean into really ambitious research in some senses. And I do wonder if now Redwood has overcorrected and w... | That’s the neuralese decoding research direction. I’ve actually written up a project proposal on this that’s publicly available and we could link in the show notes. It’s called [“Decoding Opaque Reasoning in Current Models.”](https://docs.google.com/document/d/1r0grvjYW89xrFgGtUENY0rSaEs--LIVJi0E-TsmPzAc/edit)
That’s ... | {"chunk_index": 216, "chunk_kind": "mid", "coauthored": false, "coauthors": [], "format": "continuation", "para_range": [155, 165], "post_id": "p4iJpumHt6Ay9KnXT"} | continuation |
You are continuing a LessWrong post by Ryan Greenblatt.
Title: The inaugural Redwood Research podcast
Opening:
[Earlier portions of the post are elided.]
Recent paragraphs:
**Ryan**: Anyway, I think there’s a bunch of work potentially in this space motivated by: suppose we had neuralese models, what would the straig... | And then I’ll do a bonus one quickly: I think in general, high-level interpretability on how do LLMs make decisions in pursuit of goals—what is the internal mechanism for that—could be pretty interesting. There’s a bunch of intervention-style experiments that don’t depend much on low-level understanding. I think we can... | {"chunk_index": 217, "chunk_kind": "mid", "coauthored": false, "coauthors": [], "format": "continuation", "para_range": [164, 181], "post_id": "p4iJpumHt6Ay9KnXT"} | continuation |
You are continuing a LessWrong post by Ryan Greenblatt.
Title: The inaugural Redwood Research podcast
Opening:
[Earlier portions of the post are elided.]
Recent paragraphs:
Another one of these: I’ve wanted to pay someone for a while to get really good at answering the question of—if you have two features of text, f... | **Buck**: Yeah.
**Ryan**: And it’s like, actually this is worse. We know this is worse. And in fact people have researched the optimal number of steps to do in a training run and it turns out there’s some criticality. My understanding is—this may be false—but as you scale, the number of steps you need is roughly const... | {"chunk_index": 218, "chunk_kind": "mid", "coauthored": false, "coauthors": [], "format": "continuation", "para_range": [180, 202], "post_id": "p4iJpumHt6Ay9KnXT"} | continuation |
You are continuing a LessWrong post by Ryan Greenblatt.
Title: The inaugural Redwood Research podcast
Opening:
[Earlier portions of the post are elided.]
Recent paragraphs:
**Buck**: Yeah. And to be clear, this is to be within epsilon of optimality.
**Ryan**: Yeah, this is like presumably more steps is always bette... | *Ryan’s never-published mech interp research*
**Buck**: Anyway, to continue the narrative: it was, I think, a big update that we made about interpretability. It was late 2022, and we had, through a couple of different mechanisms, come up with better ways of understanding what’s happening in neural nets and understandi... | {"chunk_index": 219, "chunk_kind": "mid", "coauthored": false, "coauthors": [], "format": "continuation", "para_range": [201, 215], "post_id": "p4iJpumHt6Ay9KnXT"} | continuation |
You are continuing a LessWrong post by Ryan Greenblatt.
Title: The inaugural Redwood Research podcast
Opening:
[Earlier portions of the post are elided.]
Recent paragraphs:
**Buck**: Yeah, I think what we should have done is not do interpretability, basically. I think we should have gone harder on the futurism stuff... | **Buck**: The thing which was novel about this research—and I think people haven’t really tried to do afterwards, either because it doesn’t work very well—is: pick some big part of the model’s behavior, like how it predicts “is” versus “was,” and then just try to understand what’s going on. And it’s very much—you refer... | {"chunk_index": 220, "chunk_kind": "mid", "coauthored": false, "coauthors": [], "format": "continuation", "para_range": [214, 229], "post_id": "p4iJpumHt6Ay9KnXT"} | continuation |
You are continuing a LessWrong post by Ryan Greenblatt.
Title: The inaugural Redwood Research podcast
Opening:
[Earlier portions of the post are elided.]
Recent paragraphs:
**Ryan**: And a lot of random stuff is going to go into that because anything that would be helpful for predicting “is” versus “was” would be re... | **Buck**: Yeah.
**Ryan**: And there’s also—I mean, biology also—a key thing is proteins are extremely natural because there’s heavy specialization where you just have ribosomes, that’s what you have, and that’s how a ton of the stuff gets manufactured. Whereas the corresponding level of abstraction would be a matrix m... | {"chunk_index": 221, "chunk_kind": "mid", "coauthored": false, "coauthors": [], "format": "continuation", "para_range": [228, 243], "post_id": "p4iJpumHt6Ay9KnXT"} | continuation |
You are continuing a LessWrong post by Ryan Greenblatt.
Title: The inaugural Redwood Research podcast
Opening:
[Earlier portions of the post are elided.]
Recent paragraphs:
**Ryan**: Yeah. And I do feel like there have been huge goalpost shifts in interpretability, at least relative to public communication about it.... | **Ryan**: Yep.
**Buck**: Yeah. What else is interesting there? I think a general phenomenon about Anthropic is: Anthropic as an organization cares more about seeming safe, and staff at Anthropic care more about having reasonable things to say about their AI safety plans and about how they’re going to not impose huge a... | {"chunk_index": 222, "chunk_kind": "mid", "coauthored": false, "coauthors": [], "format": "continuation", "para_range": [242, 253], "post_id": "p4iJpumHt6Ay9KnXT"} | continuation |
You are continuing a LessWrong post by Ryan Greenblatt.
Title: The inaugural Redwood Research podcast
Opening:
[Earlier portions of the post are elided.]
Recent paragraphs:
**Ryan**: Yeah, I think it’s a little bit complicated. It’s more like—it depends on who they were talking to and who the person is. But I think ... | *Back to the narrative*
**Buck**: Okay, why don’t we get back to the interpretability stuff? An event which occurred in the course of our increasing interp pessimism in something like December 2022 was: we went off and thought about whether mechanistic interpretability was actually promising in the way that we had bel... | {"chunk_index": 223, "chunk_kind": "mid", "coauthored": false, "coauthors": [], "format": "continuation", "para_range": [252, 265], "post_id": "p4iJpumHt6Ay9KnXT"} | continuation |
You are continuing a LessWrong post by Ryan Greenblatt.
Title: The inaugural Redwood Research podcast
Opening:
[Earlier portions of the post are elided.]
Recent paragraphs:
I’m not sure if I’ve articulated this all very clearly.
*Back to the narrative*
**Buck**: Okay, why don’t we get back to the interpretability ... | **Buck**: We were like, “We always need to have good baselines. We always need to do the simplest thing first.” And at this point we were also saying we should think carefully about which properties precisely of the future situation will mean that we should be pessimistic about behavioral methods or black-box methods. ... | {"chunk_index": 224, "chunk_kind": "mid", "coauthored": false, "coauthors": [], "format": "continuation", "para_range": [264, 276], "post_id": "p4iJpumHt6Ay9KnXT"} | continuation |
You are continuing a LessWrong post by Ryan Greenblatt.
Title: The inaugural Redwood Research podcast
Opening:
[Earlier portions of the post are elided.]
Recent paragraphs:
**Ryan**: Some projects we were thinking about at the time: one thing we were looking at around this time, maybe a little earlier, was runtime b... | **Buck**: Going down a bunch of hilarious rabbit holes in retrospect, like doing a bunch of Tom Davidson-style takeoff speed analysis or intelligence explosion analysis and getting into some of the microeconomics there, which I don’t think anything happened with that work until Tom made a better version of it later. Bu... | {"chunk_index": 225, "chunk_kind": "mid", "coauthored": false, "coauthors": [], "format": "continuation", "para_range": [275, 294], "post_id": "p4iJpumHt6Ay9KnXT"} | continuation |
You are continuing a LessWrong post by Ryan Greenblatt.
Title: The inaugural Redwood Research podcast
Opening:
[Earlier portions of the post are elided.]
Recent paragraphs:
**Buck**: Yeah, like four months or something. And this was an average of two or three people working on it.
**Ryan**: Yeah. It seems like it c... | **Buck**: Okay, all right. That is definitely good to know. Definitely it’s a little unapproachable. I think the basic issue with working with Ryan is that he has a much larger working memory than a lot of people. And so for him, the optimal strategy often is to just have 30 cursed things happening at the same time in ... | {"chunk_index": 226, "chunk_kind": "mid", "coauthored": false, "coauthors": [], "format": "continuation", "para_range": [293, 307], "post_id": "p4iJpumHt6Ay9KnXT"} | continuation |
You are continuing a LessWrong post by Ryan Greenblatt.
Title: The inaugural Redwood Research podcast
Opening:
[Earlier portions of the post are elided.]
Recent paragraphs:
**Ryan**: I mean, it’s just a normal Vim plugin.
**Buck**: Sorry, by someone else.
**Ryan**: Yeah.
**Buck**: Okay, all right. That is definit... | **Ryan**: I was interested in looking at: if you prompt models to do a competent job of pursuing objectives you give them, do they just naturally generalize? There’s some simple claim which is: obviously models should do the alignment faking thing as a special case of pursuing goals in the world. If we set up models to... | {"chunk_index": 227, "chunk_kind": "mid", "coauthored": false, "coauthors": [], "format": "continuation", "para_range": [306, 326], "post_id": "p4iJpumHt6Ay9KnXT"} | continuation |
You are continuing a LessWrong post by Ryan Greenblatt.
Title: The inaugural Redwood Research podcast
Opening:
[Earlier portions of the post are elided.]
Recent paragraphs:
**Buck**: I think this is undefeatedly interesting. This was substantially before o1 came out, and you had stumbled upon the idea—perhaps obviou... | **Ryan**: Do you want to talk about maybe some of the stuff that was going on in parallel with the alignment faking in large language models paper—what ended up being the [Ctrl-Z paper](https://arxiv.org/abs/2504.10374)?
**Buck**: Yeah. A big thing which was happening in parallel with the alignment faking paper was th... | {"chunk_index": 228, "chunk_kind": "mid", "coauthored": false, "coauthors": [], "format": "continuation", "para_range": [325, 342], "post_id": "p4iJpumHt6Ay9KnXT"} | continuation |
You are continuing a LessWrong post by Ryan Greenblatt.
Title: The inaugural Redwood Research podcast
Opening:
[Earlier portions of the post are elided.]
Recent paragraphs:
**Ryan**: Slightly smart. Maybe the AIs can come up with slightly smart ways of making it happen.
**Buck**: And basically a lot of the dynamics... | **Ryan**: This is like a performance improvement plan.
**Buck**: Yeah, stuff like that. So we ended up shrinking moderately by the beginning of 2023, and we ended up shrinking further through 2023 until we ended up having four full-time staff at the point where we wrote the control paper. This was again some mixture o... | {"chunk_index": 229, "chunk_kind": "mid", "coauthored": false, "coauthors": [], "format": "continuation", "para_range": [341, 352], "post_id": "p4iJpumHt6Ay9KnXT"} | continuation |
You are continuing a LessWrong post by Ryan Greenblatt.
Title: How do we (more) safely defer to AIs?
Opening:
As AI systems get more capable, it becomes increasingly uncompetitive and infeasible to avoid deferring to AIs on increasingly many decisions. Further, once systems are sufficiently capable, [control becomes ... | Our ultimate goal is to have the AIs we defer to resolve risks associated with powerful AI systems (and potentially other serious risks that happen to occur at this time) while preserving option value and keeping humans in control of longer-run values-loaded decision making (e.g., how Earth should be governed in the lo... | {"chunk_index": 230, "chunk_kind": "head", "coauthored": true, "coauthors": ["Julian Stastny"], "format": "continuation", "para_range": [10, 26], "post_id": "vjAM7F8vMZS7oRrrh"} | continuation |
You are continuing a LessWrong post by Ryan Greenblatt.
Title: How do we (more) safely defer to AIs?
Opening:
[Earlier portions of the post are elided.]
Recent paragraphs:
## Recursive self-improvement of alignment and wisdom: the hope for a Basin of Good Deference
A key hope is that the initial AIs we defer to wil... | We can break down deference-goodness into an alignment/elicitation component and a capability profile component:
* **Broad alignment**: the AIs need to be aligned and well elicited enough that they faithfully pursue our interests even on massive, open-ended, and conceptually-loaded tasks (with poor feedback loops) whe... | {"chunk_index": 231, "chunk_kind": "mid", "coauthored": true, "coauthors": ["Julian Stastny"], "format": "continuation", "para_range": [25, 32], "post_id": "vjAM7F8vMZS7oRrrh"} | continuation |
You are continuing a LessWrong post by Ryan Greenblatt.
Title: How do we (more) safely defer to AIs?
Opening:
[Earlier portions of the post are elided.]
Recent paragraphs:
This requires that at each point after deference, our AIs:
* Aren't seriously misaligned: They remain corrigible to the intended group/structure... | Another possibility—which isn't a problem we can solve by doing a better job with the initial AIs we defer to—is that deference-goodness is just fundamentally persistently hard to advance relative to capabilities such that the ongoing amount of effort required to advance deference-goodness is too high. In general, a ke... | {"chunk_index": 232, "chunk_kind": "mid", "coauthored": true, "coauthors": ["Julian Stastny"], "format": "continuation", "para_range": [31, 41], "post_id": "vjAM7F8vMZS7oRrrh"} | continuation |
You are continuing a LessWrong post by Ryan Greenblatt.
Title: How do we (more) safely defer to AIs?
Opening:
[Earlier portions of the post are elided.]
Recent paragraphs:
Ultimately, this deference-goodness tax bottoms out in terms of an amount of additional lead time (relative to exogenous risks) that AIs we defer... | Beyond scheming, I expect that it's easier to make AIs which are less capable (but still at least comparably capable to human experts) more aligned in general (e.g. more likely to actually try and pursue our interests even on very hard to check tasks) as these less capable AIs are generally easier to oversee on a given... | {"chunk_index": 233, "chunk_kind": "mid", "coauthored": true, "coauthors": ["Julian Stastny"], "format": "continuation", "para_range": [40, 50], "post_id": "vjAM7F8vMZS7oRrrh"} | continuation |
You are continuing a LessWrong post by Ryan Greenblatt.
Title: How do we (more) safely defer to AIs?
Opening:
[Earlier portions of the post are elided.]
Recent paragraphs:
If we ended up finding that a substantially higher level of capabilities would actually improve deference-goodness (while still making scheming m... | This does depend on AIs running substantially faster than humans (and being more numerous) but I expect this will be true in practice (if it isn't true, then ideally we'd optimize for it being true because faster and more numerous AIs at a lower level of qualitative capability would be safer).
More generally, I think ... | {"chunk_index": 234, "chunk_kind": "mid", "coauthored": true, "coauthors": ["Julian Stastny"], "format": "continuation", "para_range": [49, 60], "post_id": "vjAM7F8vMZS7oRrrh"} | continuation |
You are continuing a LessWrong post by Ryan Greenblatt.
Title: How do we (more) safely defer to AIs?
Opening:
[Earlier portions of the post are elided.]
Recent paragraphs:
This does depend on AIs running substantially faster than humans (and being more numerous) but I expect this will be true in practice (if it isn'... | Given that current human experts currently disagree strongly about important aspects of the situation and what should be done, do epistemics at the level of top human experts suffice? My guess is that the answer is yes, though we may need to have the AIs ensemble together different epistemic strategies to diversify. Pa... | {"chunk_index": 235, "chunk_kind": "mid", "coauthored": true, "coauthors": ["Julian Stastny"], "format": "continuation", "para_range": [59, 68], "post_id": "vjAM7F8vMZS7oRrrh"} | continuation |
You are continuing a LessWrong post by Ryan Greenblatt.
Title: How do we (more) safely defer to AIs?
Opening:
[Earlier portions of the post are elided.]
Recent paragraphs:
Given that current human experts currently disagree strongly about important aspects of the situation and what should be done, do epistemics at t... | I've been using the term "alignment" and I'll focus on approaches based on actually robustly aligning AI systems, but even when fully deferring some types of not-fully-worst-case misalignment problems could in principle be resolved by mixing control and alignment or possibly with (white-box) control-style approaches. F... | {"chunk_index": 236, "chunk_kind": "mid", "coauthored": true, "coauthors": ["Julian Stastny"], "format": "continuation", "para_range": [67, 78], "post_id": "vjAM7F8vMZS7oRrrh"} | continuation |
You are continuing a LessWrong post by Ryan Greenblatt.
Title: How do we (more) safely defer to AIs?
Opening:
[Earlier portions of the post are elided.]
Recent paragraphs:
Even if we had very good and cheap behavioral tests (as in, these test beds are almost perfectly analogous, aren't themselves risky to run, and w... | It's also plausible we'll want to augment the AI's ability to specifically do the work we want to hand off. This could look like generally trying to improve the AI's ability to work in a given field without having a good model of what underlying capabilities/heuristics this might be providing (e.g. via RL in safety res... | {"chunk_index": 237, "chunk_kind": "mid", "coauthored": true, "coauthors": ["Julian Stastny"], "format": "continuation", "para_range": [77, 89], "post_id": "vjAM7F8vMZS7oRrrh"} | continuation |
You are continuing a LessWrong post by Ryan Greenblatt.
Title: How do we (more) safely defer to AIs?
Opening:
[Earlier portions of the post are elided.]
Recent paragraphs:
One concern is that sufficient capabilities for deference will necessarily imply that the AI is (or easily could be) extremely superhuman on some... | Here's a more detailed guess at what our alignment target should be like:
* The AIs are corrigible to human modification and shutdown through the official process/channel[^29] and don't aim to obstruct this in any way. The AIs don't attempt to seize power for their own ends.
* The AIs (generally speaking) try to: obey... | {"chunk_index": 238, "chunk_kind": "mid", "coauthored": true, "coauthors": ["Julian Stastny"], "format": "continuation", "para_range": [88, 96], "post_id": "vjAM7F8vMZS7oRrrh"} | continuation |
You are continuing a LessWrong post by Ryan Greenblatt.
Title: How do we (more) safely defer to AIs?
Opening:
[Earlier portions of the post are elided.]
Recent paragraphs:
Keep in mind that this setup means we're directly instructing (or possibly training) the AIs to pursue big open-ended projects such as "figure ou... | One strong type of alignment is the AI using all of its abilities and affordances (that it can use) toward achieving the intended aims, including abilities which are very far from things we could directly supervise during training, e.g., using its introspective understanding of its own preferences. This is probably not... | {"chunk_index": 239, "chunk_kind": "mid", "coauthored": true, "coauthors": ["Julian Stastny"], "format": "continuation", "para_range": [95, 104], "post_id": "vjAM7F8vMZS7oRrrh"} | continuation |
You are continuing a LessWrong post by Ryan Greenblatt.
Title: How do we (more) safely defer to AIs?
Opening:
[Earlier portions of the post are elided.]
Recent paragraphs:
One strong type of alignment is the AI using all of its abilities and affordances (that it can use) toward achieving the intended aims, including... | One open question is what epistemics target suffices. A common objection to deferring to AIs is that AI safety researchers disagree massively, so correspondingly, it must be really hard to get epistemics right (otherwise we'd agree). I think I mostly don't buy this objection, though I do agree that making errors on tri... | {"chunk_index": 240, "chunk_kind": "mid", "coauthored": true, "coauthors": ["Julian Stastny"], "format": "continuation", "para_range": [103, 113], "post_id": "vjAM7F8vMZS7oRrrh"} | continuation |
You are continuing a LessWrong post by Ryan Greenblatt.
Title: How do we (more) safely defer to AIs?
Opening:
[Earlier portions of the post are elided.]
Recent paragraphs:
Of course, we'd also attempt to make the AI superhuman or at least match the best humans on epistemic properties that are pretty clearly good (or... | Given that we'll have imperfect testing with some limitations, it will be important to have ways of iterating on methods which are less likely to overfit to our tests. In the sections below, I'll discuss more specific proposals for improving the capability profile and broad alignment of our systems, but at a high level... | {"chunk_index": 241, "chunk_kind": "mid", "coauthored": true, "coauthors": ["Julian Stastny"], "format": "continuation", "para_range": [112, 121], "post_id": "vjAM7F8vMZS7oRrrh"} | continuation |
You are continuing a LessWrong post by Ryan Greenblatt.
Title: How do we (more) safely defer to AIs?
Opening:
[Earlier portions of the post are elided.]
Recent paragraphs:
My overall sense is that this behavioral testing will generally be hard. It will probably be a huge mess if we're extremely rushed and need to do... | Work tests are likely a key part of our approach, but due to our limited oversight ability in the domain, we might only be able to test so far. We can look at how the AIs solve smaller problems (and where we feel better about our ability to oversee their results).
While work tests work decently well in the human case ... | {"chunk_index": 242, "chunk_kind": "mid", "coauthored": true, "coauthors": ["Julian Stastny"], "format": "continuation", "para_range": [120, 129], "post_id": "vjAM7F8vMZS7oRrrh"} | continuation |
You are continuing a LessWrong post by Ryan Greenblatt.
Title: How do we (more) safely defer to AIs?
Opening:
[Earlier portions of the post are elided.]
Recent paragraphs:
Some types of misalignment are easy to detect with cheap tests, e.g., some types of reward seeking propensities should be easy to detect. So we s... | One particular type of problematic generalization is that we might end up with an AI which either explicitly reasons about getting high performance on things we might test (and not in a way which fits to the exact training distribution, rather it tries to cover whatever the best tests humans might have been able to con... | {"chunk_index": 243, "chunk_kind": "mid", "coauthored": true, "coauthors": ["Julian Stastny"], "format": "continuation", "para_range": [128, 139], "post_id": "vjAM7F8vMZS7oRrrh"} | continuation |
You are continuing a LessWrong post by Ryan Greenblatt.
Title: How do we (more) safely defer to AIs?
Opening:
[Earlier portions of the post are elided.]
Recent paragraphs:
If we get an AI which has heuristics that end up doing well on our tests (but not in practice) because these heuristics also gamed our training, ... | Here are some additional more speculative hopes for behavioral testing:
* In some cases, we can set up situations where we happen to be able to verify some output but producing a good output still requires capabilities/propensities which are applicable to some very hard-to-check task. In the case where we directly tra... | {"chunk_index": 244, "chunk_kind": "mid", "coauthored": true, "coauthors": ["Julian Stastny"], "format": "continuation", "para_range": [138, 146], "post_id": "vjAM7F8vMZS7oRrrh"} | continuation |
You are continuing a LessWrong post by Ryan Greenblatt.
Title: How do we (more) safely defer to AIs?
Opening:
[Earlier portions of the post are elided.]
Recent paragraphs:
We should make sure to directly look at how the AI relates to acausal trade and decision theory and make sure its perspective seems reasonable or... | How can we resolve these issues? I'll discuss some of the more straightforward approaches focused on this problem:
* **Improve feedback**: The most obvious approach will be to directly improve our feedback quality or how feedback is used in training. This isn't applicable if the issue isn't due to bad feedback of cour... | {"chunk_index": 245, "chunk_kind": "mid", "coauthored": true, "coauthors": ["Julian Stastny"], "format": "continuation", "para_range": [145, 148], "post_id": "vjAM7F8vMZS7oRrrh"} | continuation |
You are continuing a LessWrong post by Ryan Greenblatt.
Title: How do we (more) safely defer to AIs?
Opening:
[Earlier portions of the post are elided.]
Recent paragraphs:
* **Improve feedback**: The most obvious approach will be to directly improve our feedback quality or how feedback is used in training. This isn'... | It's worth noting there are two somewhat different concerns with learning problematic preferences from RL environments:
1. Environments which don't look much like our deployment context (but which are used to instill relevant capabilities or alignment properties) result in problematic propensities which transfer.
2. W... | {"chunk_index": 246, "chunk_kind": "mid", "coauthored": true, "coauthors": ["Julian Stastny"], "format": "continuation", "para_range": [147, 157], "post_id": "vjAM7F8vMZS7oRrrh"} | continuation |
You are continuing a LessWrong post by Ryan Greenblatt.
Title: How do we (more) safely defer to AIs?
Opening:
[Earlier portions of the post are elided.]
Recent paragraphs:
It's worth noting there are two somewhat different concerns with learning problematic preferences from RL environments:
1. Environments which do... | First, relatively more prosaic and straightforwardly implementable approaches (some of these approaches overlap with things I've discussed above):
* **Build lots of alignment specialized environments**: An important part of getting the relevant alignment properties might be building a bunch of environments focused on ... | {"chunk_index": 247, "chunk_kind": "mid", "coauthored": true, "coauthors": ["Julian Stastny"], "format": "continuation", "para_range": [156, 159], "post_id": "vjAM7F8vMZS7oRrrh"} | continuation |
You are continuing a LessWrong post by Ryan Greenblatt.
Title: How do we (more) safely defer to AIs?
Opening:
[Earlier portions of the post are elided.]
Recent paragraphs:
* **Build lots of alignment specialized environments**: An important part of getting the relevant alignment properties might be building a bunch ... | And here are some more ambitious proposals (discussed in minimal detail):
* **Ensure legible chain-of-thought and directly steer thoughts**. If we could understand and intervene on the "thoughts" the AI has, that could help substantially resolve misalignment issues (though some issues might operate at a lower level in... | {"chunk_index": 248, "chunk_kind": "mid", "coauthored": true, "coauthors": ["Julian Stastny"], "format": "continuation", "para_range": [158, 166], "post_id": "vjAM7F8vMZS7oRrrh"} | continuation |
You are continuing a LessWrong post by Ryan Greenblatt.
Title: How do we (more) safely defer to AIs?
Opening:
[Earlier portions of the post are elided.]
Recent paragraphs:
Remember that I'm focusing on alignment approaches, but some control style approaches could still be helpful after deference, especially as we mi... | The basic strategies will be:
* Try to construct RL environments or synthetic data which we think hit the relevant capabilities.
* There are some domains which are loaded on tricky epistemic / conceptual skills (that might partially transfer) and which are possible to easily score. For instance, forecasting, decisio... | {"chunk_index": 249, "chunk_kind": "mid", "coauthored": true, "coauthors": ["Julian Stastny"], "format": "continuation", "para_range": [165, 179], "post_id": "vjAM7F8vMZS7oRrrh"} | continuation |
You are continuing a LessWrong post by Ryan Greenblatt.
Title: How do we (more) safely defer to AIs?
Opening:
[Earlier portions of the post are elided.]
Recent paragraphs:
If the paradigm involves things other than mostly normal machine learning methods (e.g., earlier AIs constructing circuits by hand), then our app... | Doing a good job with deference involves creating AIs that have a good understanding of the situation, including understanding truths that people at the AI company probably don't want to hear and/or don't want to spread. Most centrally, that the AI company has been very reckless or is otherwise doing extremely dangerou... | {"chunk_index": 250, "chunk_kind": "mid", "coauthored": true, "coauthors": ["Julian Stastny"], "format": "continuation", "para_range": [178, 193], "post_id": "vjAM7F8vMZS7oRrrh"} | continuation |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.