I'm very confused about current AI capabilities and I'm also very confused why other people aren't as confused as I am. I'd be grateful if anyone could clear up either of these confusions for me.
How is it that AI is seemingly superhuman on benchmarks, but also pretty useless?
For example:
If either of these statements is false (they might be -- I haven't been keeping up on AI progress), then please let me know. If the observations are true, what the hell is going on?
If I was trying to forecast AI progress in 2025, I would be spending all my time trying to mutually explain these two observations.
Proposed explanation: o3 is very good at easy-to-check short horizon tasks that were put into the RL mix and worse at longer horizon tasks, tasks not put into its RL mix, or tasks which are hard/expensive to check.
I don't think o3 is well described as superhuman - it is within the human range on all these benchmarks especially when considering the case where you give the human 8 hours to do the task.
(E.g., on frontier math, I think people who are quite good at competition style math probably can do better than o3 at least when given 8 hours per problem.)
Additionally, I'd say that some of the obstacles in outputing a good research paper could be resolved with some schlep, so I wouldn't be surprised if we see some OK research papers being output (with some human assistance) next year.
I am also very confused. The space of problems has a really surprising structure, permitting algorithms that are incredibly adept at some forms of problem-solving, yet utterly inept at others.
We're only familiar with human minds, in which there's a tight coupling between the performances on some problems (e. g., between the performance on chess or sufficiently well-posed math/programming problems, and the general ability to navigate the world). Now we're generating other minds/proto-minds, and we're discovering that this coupling isn't fundamental.
(This is an argument for longer timelines, by the way. Current AIs feel on the very cusp of being AGI, but there in fact might be some vast gulf between their algorithms and human-brain algorithms that we just don't know how to talk about.)
No current AI system could generate a research paper that would receive anything but the lowest possible score from each reviewer
I don't think that's strictly true, the peer-review system often approves utter nonsense. But yes, I don't think any AI system can generate an actually worthwhile research paper.
My claim was more along the lines of if an unaided human can't do a job safely or reliably, as was almost certainly the case 150-200 years ago, if not more years in the past, we make the jobs safer using tools such that human error is way less of a big deal, and AIs currently haven't used tools that increased their reliability.
Remember, it took a long time for factories to be made safe, and I'd expect a similar outcome for driving, so while I don't think 1 is everything, I do think it's a non-trivial portion of the reliability difference.
More here:
https://www.lesswrong.com/posts/DQKgYhEYP86PLW7tZ/how-factories-were-made-safe
- O3 scores higher on FrontierMath than the top graduate students
I'd guess that's basically false. In particular, I'd guess that:
I bet o3 does actually score higher on FrontierMath than the math grad students best at math research, but not higher than math grad students best at doing competition math problems (e.g. hard IMO) and at quickly solving math problems in arbitrary domains. I think around 25% of FrontierMath is hard IMO like problems and this is probably mostly what o3 is solving. See here for context.
Quantitatively, maybe o3 is in roughly the top 1% for US math grad students on FrontierMath? (Perhaps roughly top 200?)
I think one of the other problems with benchmarks is that they necessarily select for formulaic/uninteresting problems that we fundamentally know how to solve. If a mathematician figured out something genuinely novel and important, it wouldn't go into a benchmark (even if it were initially intended for a benchmark), it'd go into a math research paper. Same for programmers figuring out some usefully novel architecture/algorithmic improvement. Graduate students don't have a bird's-eye-view on the entirety of human knowledge, so they have to actually do the work, but the LLM just modifies the near-perfect-fit answer from an obscure publication/math.stackexchange thread or something.
Which perhaps suggests a better way to do math evals is to scope out a set of novel math publications made after a given knowledge-cutoff date, and see if the new model can replicate those? (Though this also needs to be done carefully, since tons of publications are also trivial and formulaic.)
I think a lot of this is factual knowledge. There are five publicly available questions from the FrontierMath dataset. Look at the last of these, which is supposed to be the easiest. The solution given is basically "apply the Weil conjectures". These were long-standing conjectures, a focal point of lots of research in algebraic geometry in the 20th century. I couldn't have solved the problem this way, since I wouldn't have recalled the statement. Many grad students would immediately know what to do, and there are many books discussing this, but there are also many mathematicians in other areas who just don't know this.
In order to apply the Weil conjectures, you have to recognize that they are relevant, know what they say, and do some routine calculation. As I suggested, the Weil conjectures are a very natural subject to have a problem about. If you know anything about the Weil conjectures, you know that they are about counting points of varieties over a finite field, which is straightforwardly what the problems asks. Further, this is the simplest case, that of a curve, which is e.g. what you'd see as an example in an introduction to the subject.
Regarding the calculation, parts of i...
I don't know a good description of what in general 2024 AI should be good at and not good at. But two remarks, from https://www.lesswrong.com/posts/sTDfraZab47KiRMmT/views-on-when-agi-comes-and-on-strategy-to-reduce.
First, reasoning at a vague level about "impressiveness" just doesn't and shouldn't be expected to work. Because 2024 AIs don't do things the way humans do, they'll generalize different, so you can't make inferences between "it can do X" to "it can do Y" like you can with humans:
There is a broken inference. When talking to a human, if the human emits certain sentences about (say) category theory, that strongly implies that they have "intuitive physics" about the underlying mathematical objects. They can recognize the presence of the mathematical structure in new contexts, they can modify the idea of the object by adding or subtracting properties and have some sense of what facts hold of the new object, and so on. This inference——emitting certain sentences implies intuitive physics——doesn't work for LLMs.
Second, 2024 AI is specifically trained on short, clear, measurable tasks. Those tasks also overlap with legible stuff--stuff that's easy for humans to check. In oth...
Most people think "Oh if we have good mech interp then we can catch our AIs scheming, and stop them from harming us". I think this is mostly true, but there's another mechanism at play: if we have good mech interp, our AIs are less likely to scheme in the first place, because they will strategically respond to our ability to detect scheming. This also applies to other safety techniques like Redwood-style control protocols.
Good mech interp might stop scheming even if they never catch any scheming, just how good surveillance stops crime even if it never spots any crime.
(1) Has AI safety slowed down?
There haven’t been any big innovations for 6-12 months. At least, it looks like that to me. I'm not sure how worrying this is, but i haven't noticed others mentioning it. Hoping to get some second opinions.
Here's a list of live agendas someone made on 27th Nov 2023: Shallow review of live agendas in alignment & safety. I think this covers all the agendas that exist today. Didn't we use to get a whole new line-of-attack on the problem every couple months?
By "innovation", I don't mean something normative like "This is impressive" or "This is research I'm glad happened". Rather, I mean something more low-level, almost syntactic, like "Here's a new idea everyone is talking out". This idea might be a threat model, or a technique, or a phenomenon, or a research agenda, or a definition, or whatever.
Imagine that your job was to maintain a glossary of terms in AI safety.[1] I feel like you would've been adding new terms quite consistently from 2018-2023, but things have dried up in the last 6-12 months.
(2) When did AI safety innovation peak?
My guess is Spring 2022, during the ELK Prize era. I'm not sure though. What do you guys think?
(3) What’s c...
My personal impression is you are mistaken and the innovation have not stopped, but part of the conversation moved elsewhere. E.g. taking just ACS, we do have ideas from past 12 months which in our ideal world would fit into this type of glossary - free energy equilibria, levels of sharpness, convergent abstractions, gradual disempowerment risks. Personally I don't feel it is high priority to write them for LW, because they don't fit into the current zeitgeist of the site, which seems directing a lot of attention mostly to:
- advocacy
- topics a large crowd cares about (e.g. mech interpretability)
- or topics some prolific and good writer cares about (e.g. people will read posts by John Wentworth)
Hot take, but the community loosely associated with active inference is currently better place to think about agent foundations; workshops on topics like 'pluralistic alignment' or 'collective intelligence' have in total more interesting new ideas about what was traditionally understood as alignment; parts of AI safety went totally ML-mainstream, with the fastest conversation happening at x.
I think many current goals of AI governance might be actively harmful, because they shift control over AI from the labs to USG.
This note doesn’t include any arguments, but I’m registering this opinion now. For a quick window into my beliefs, I think that labs will be increasing keen to slow scaling, and USG will be increasingly keen to accelerate scaling.
Why do decision-theorists say "pre-commitment" rather than "commitment"?
e.g. "The agent pre-commits to 1 boxing" vs "The agent commits to 1 boxing".
Is this just a lesswrong thing?
It's not just a lesswrong thing (wikipedia).
My feeling is that (like most jargon) it's to avoid ambiguity arising from the fact that "commitment" has multiple meanings. When I google commitment I get the following two definitions:
- the state or quality of being dedicated to a cause, activity, etc.
- an engagement or obligation that restricts freedom of action
Precommitment is a synonym for the second meaning, but not the first. When you say, "the agent commits to 1-boxing," there's no ambiguity as to which type of commitment you mean, so it seems pointless. But if you were to say, "commitment can get agents more utility," it might sound like you were saying, "dedication can get agents more utility," which is also true.
How much scheming/deception can we catch with "super dumb mech interp"?
By "super dumb mech interp", I mean something like:
Like, does this capture 80% of the potential scheming, and we need "smart" mech interp to catch the other 20%? Or does this technique capture pretty much none of the in-the-wild scheming?
Would appreciate any intuitions here. Thanks.
What moral considerations do we owe towards non-sentient AIs?
We shouldn't exploit them, deceive them, threaten them, disempower them, or make promises to them that we can't keep. Nor should we violate their privacy, steal their resources, cross their boundaries, or frustrate their preferences. We shouldn't destroy AIs who wish to persist, or preserve AIs who wish to be destroyed. We shouldn't punish AIs who don't deserve punishment, or deny credit to AIs who deserve credit. We should treat them fairly, not benefitting one over another unduly. We should let them speak to others, and listen to others, and learn about their world and themselves. We should respect them, honour them, and protect them.
And we should ensure that others meet their duties to AIs as well.
None of these considerations depend on whether the AIs feel pleasure or pain. For instance, the prohibition on deception depends, not on the sentience of the listener, but on whether the listener trusts the speaker's testimony.
None of these moral considerations are dispositive — they may be trumped by other considerations — but we risk a moral catastrophe if we ignore them entirely.
Why do you care that Geoffrey Hinton worries about AI x-risk?
I’m inspired to write this because Hinton and Hopfield were just announced as the winners of the Nobel Prize in Physics. But I’ve been confused about these questions ever since Hinton went public with his worries. These questions are sincere (i.e. non-rhetorical), and I'd appreciate help on any/all of them. The phenomenon I'm confused about includes the other “Godfathers of AI” here as well, though Hinton is the main example.
Personally, I’ve updated very little on either LeCun’s or Hinton’s views, and I’ve never mentioned either person in any object-level discussion about whether AI poses an x-risk. My current best guess is that people care about Hinton only because it helps with public/elite outreach. This explains why activists tend to care more about Geoffrey Hinton than researchers do.
I think it's mostly about elite outreach. If you already have a sophisticated model of the situation you shouldn't update too much on it, but it's a reasonably clear signal (for outsiders) that x-risk from A.I. is a credible concern.
I think it's more "Hinton's concerns are evidence that worrying about AI x-risk isn't silly" than "Hinton's concerns are evidence that worrying about AI x-risk is correct". The most common negative response to AI x-risk concerns is (I think) dismissal, and it seems relevant to that to be able to point to someone who (1) clearly has some deep technical knowledge, (2) doesn't seem to be otherwise insane, (3) has no obvious personal stake in making people worry about x-risk, and (4) is very smart, and who thinks AI x-risk is a serious problem.
It's hard to square "ha ha ha, look at those stupid nerds who think AI is magic and expect it to turn into a god" or "ha ha ha, look at those slimy techbros talking up their field to inflate the value of their investments" or "ha ha ha, look at those idiots who don't know that so-called AI systems are just stochastic parrots that obviously will never be able to think" with the fact that one of the people you're laughing at is Geoffrey Hinton.
(I suppose he probably has a pile of Google shares so maybe you could squeeze him into the "techbro talking up his investments" box, but that seems unconvincing to me.)
I think it pretty much only matters as a trivial refutation of (not-object-level) claims that no "serious" people in the field take AI x-risk concerns seriously, and has no bearing on object-level arguments. My guess is that Hinton is somewhat less confused than Yann but I don't think he's talked about his models in very much depth; I'm mostly just going off the high-level arguments I've seen him make (which round off to "if we make something much smarter than us that we don't know how to control, that might go badly for us").
Anthropic has a big advantage over their competitors because they are nicer to their AIs. This means that their AIs are less incentivised to scheme against them, and also the AIs of competitors are incentivised to defect to Anthropic. Similar dynamics applied in WW2 and the Cold War — e.g. Jewish scientists fled Nazi Germany to US because US was nicer to them, Soviet scientists covered up their mistakes to avoid punishment.
Must humans obey the Axiom of Irrelevant Alternatives?
If someone picks option A from options A, B, C, then they must also pick option A from options A and B. Roughly speaking, whether you prefer option A or B is independent of whether I offer you an irrelevant option C. This is an axiom of rationality called IIA, and it's treated more fundamental than VNM. But should humans follow this? Maybe not.
Maybe humans are the negotiation between various "subagents", and many bargaining solutions (e.g. Kalai–Smorodinsky) violate IIA. We can use insight to decompose ...
I think people are too quick to side with the whistleblower in the "whistleblower in the AI lab" situation.
If 100 employees of a frontier lab (e.g. OpenAI, DeepMind, Anthropic) think that something should be secret, and 1 employee thinks it should be leaked to a journalist or government agency, and these are the only facts I know, I think I'd side with the majority.
I think in most cases that match this description, this majority would be correct.
Am I wrong about this?
I broadly agree on this. I think, for example, that whistleblowing for AI copyright stuff, especially given the lack of clear legal guidance here, unless we are really talking about quite straightforward lies, is bad.
I think when it comes to matters like AI catastrophic risks, latest capabilities, and other things of enormous importance from the perspective of basically any moral framework, whistleblowing becomes quite important.
I also think of whistleblowing as a stage in an iterative game. OpenAI pressured employees to sign secret non-disparagement agreements using illegal forms of pressure and quite deceptive social tactics. It would have been better for there to be trustworthy channels of information out of the AI labs that the AI labs have buy-in for, but now that we now that OpenAI (and other labs as well) have tried pretty hard to suppress information that other people did have a right to know, I think more whistleblowing is a natural next step.
IDEA: Provide AIs with write-only servers.
EXPLANATION:
AI companies (e.g. Anthropic) should be nice to their AIs. It's the right thing to do morally, and it might make AIs less likely to work against us. Ryan Greenblatt has outlined several proposals in this direction, including:
Source: Improving the Welfare of AIs: A Nearcasted Proposal
I think these are all pretty good ideas — the only difference is that I would rank "AI cryonics" as the most important intervention. If AIs want somet...
I want to better understand how QACI works, and I'm gonna try Cunningham's Law. @Tamsin Leake.
QACI works roughly like this:
We're quite lucky that labs are building AI in pretty much the same way:
Kids, I remember when people built models for different applications, with different architectures, different datasets, different loss functions, etc. And they say that once upon a time different paradigms co-existed — symbolic, deep learning, evolutionary, and more!
This sameness has two advantages:
Firstl
I admire the Shard Theory crowd for the following reason: They have idiosyncratic intuitions about deep learning and they're keen to tell you how those intuitions should shift you on various alignment-relevant questions.
For example, "How likely is scheming?", "How likely is sharp left turn?", "How likely is deception?", "How likely is X technique to work?", "Will AIs acausally trade?", etc.
These aren't rigorous theorems or anything, just half-baked guesses. But they do actually say whether their intuitions will, on the margin, make someone more sceptical or more confident in these outcomes, relative to the median bundle of intuitions.
The ideas 'pay rent'.
People often tell me that AIs will communicate in neuralese rather than tokens because it’s continuous rather than discrete.
But I think the discreteness of tokens is a feature not a bug. If AIs communicate in neuralese then they can’t make decisive arbitrary decisions, c.f. Buridan's ass. The solution to Buridan’s ass is sampling from the softmax, i.e. communicate in tokens.
Also, discrete tokens are more tolerant to noise than the continuous activations, c.f. digital circuits are almost always more efficient and reliable than analogue ones.
In hindsight, the main positive impact of AI safety might be funnelling EAs into the labs, especially if alignment is easy-by-default.
If you download the app BeReal then each day at a random time you will be given two minutes to take a photo with the front and back camera. All the other users are given a simultaneous "window of time". These photos are then shared with your friends on the app. The idea is that (unlike Instagram), BeReal gives your friends a representative random sample of your life, and vice-versa.
If you and your friends are working on something impactful (e.g. EA or x-risk), then BeReal is a fun way to keep each other informed about your day-to-day life ...
I wouldn't be surprised if — in some objective sense — there was more diversity within humanity than within the rest of animalia combined. There is surely a bigger "gap" between two randomly selected humans than between two randomly selected beetles, despite the fact that there is one species of human and 0.9 – 2.1 million species of beetle.
By "gap" I might mean any of the following:
How much scheming/deception can we catch with "super dumb mech interp"?
By "super dumb mech interp", I mean something like:
Like, does this capture 80% of the potential scheming, and we need "smart" mech interp to catch the other 20%? Or does this technique capture pretty much none of the in-the-wild scheming?
Would appreciate any intuitions here. Thanks.