Kaarel

In­ter­pretabil­ity: In­te­grated Gra­di­ents is a de­cent at­tri­bu­tion method

The Lo­cal In­ter­ac­tion Ba­sis: Iden­ti­fy­ing Com­pu­ta­tion­ally-Rele­vant and Sparsely In­ter­act­ing Fea­tures in Neu­ral Networks

• Kaarel 4 Apr 2024 14:24 UTC

  2 points

  0

  in reply to: Kaarel’s comment on: kh’s Shortform

  The Deep Neural Feature Ansatz

  @misc{radhakrishnan2023mechanism, title={Mechanism of feature learning in deep fully connected networks and kernel machines that recursively learn features}, author={Adityanarayanan Radhakrishnan and Daniel Beaglehole and Parthe Pandit and Mikhail Belkin}, year={2023}, url = { https://​​arxiv.org/​​pdf/​​2212.13881.pdf } }

  The ansatz from the paper

  Let $h_i\left(x\right)\in R^k$ denote the activation vector in layer $i$ on input $x\in R^d$, with the input layer being at index $i=1$, so $h_1\left(x\right)=x$. Let $W_i$ be the weight matrix after activation layer $i$. Let $f_i$ be the function that maps from the $i$th activation layer to the output. Then their Deep Neural Feature Ansatz says that $$W_i^T{W}_i\underset{\sim }{\propto }\frac{1}{|D|}\sum _{x\in D}\nabla f_i\left(h_i\right(x\left)\right)\nabla f_i\left(h_i\right(x){)}^T$$ (I’m somewhat confused here about them not mentioning the loss function at all — are they claiming this is reasonable for any reasonable loss function? Maybe just MSE? MSE seems to be the only loss function mentioned in the paper; I think they leave the loss unspecified in a bunch of places though.)

  A singular vector version of the ansatz

  Letting $W_i=U\Sigma V^T$ be a SVD of $W_i$, we note that this is equivalent to $$V{\Sigma }^2{V}^T\underset{\sim }{\propto }\frac{1}{|D|}\sum _{x\in D}\nabla f_i\left(h_i\right(x\left)\right)\nabla f_i\left(h_i\right(x){)}^T,$$ i.e., that the eigenvectors of the matrix $M$ on the RHS are the right singular vectors. By the variational characterization of eigenvectors and eigenvalues (Courant-Fischer or whatever), this is the same as saying that right singular vectors of $W_i$ are the highest orthonormal $v^{T}Mv$ directions for the matrix $M$ on the RHS. Plugging in the definition of $M$, this is equivalent to saying that the right singular vectors are the sequence of highest-variance directions of the data set of gradients $\nabla f_i\left(h_i\right(x\left)\right)$.

  (I have assumed here that the linearity is precise, whereas really it is approximate. It’s probably true though that with some assumptions, the approximate initial statement implies an approximate conclusion too? Getting approx the same vecs out probably requires some assumption about gaps in singular values being big enough, because the vecs are unstable around equality. But if we’re happy getting a sequence of orthogonal vectors that gets variances which are nearly optimal, we should also be fine without this kind of assumption. (This is guessing atm.))

  Getting rid of the $W_i$ dependence on the RHS?

  Assuming there isn’t an off-by-one error in the paper, we can pull some $W_i$ term out of the RHS maybe? This is because applying the chain rule to the Jacobians of the transitions $i\to i+1\to \text{end}$ gives $\nabla f_i\left(h_i\right(x){)}^T=\nabla f_{i+1}(h_{i+1}\left(x\right){)}^T{W}_i$, so $$\frac{1}{|D|}\sum _{x\in D}\nabla f_i\left(h_i\right(x\left)\right)\nabla f_i\left(h_i\right(x){)}^T=\frac{1}{|D|}\sum _{x\in D}{W}_i^T\nabla f_{i+1}(h_{i+1}\left(x\right)\left)\nabla f_{i+1}\right(h_{i+1}\left(x\right){)}^T{W}_i.$$

  Wait, so the claim is just $$W_i^T{W}_i\underset{\sim }{\propto }{W}_i^T\left(\sum _{x\in D}\nabla f_{i+1}\right(h_{i+1}\left(x\right)\left)\nabla f_{i+1}\right(h_{i+1}\left(x\right){)}^T)W_i$$ which, assuming $W_i$ is invertible, should be the same as ${\sum }_{x\in D}\nabla f_{i+1}\left(h_{i+1}\right(x\left)\right)\nabla f_{i+1}\left(h_{i+1}\right(x){)}^T\underset{\sim }{\propto }I$. But also, they claim that it is $W_{i+1}^T{W}_{i+1}$? Are they secretly approximating everything with identity matrices?? This doesn’t seem to be the case from their Figure 2 though.

  Oh oops I guess I forgot about activation functions here! There should be extra diagonal terms for jacobians of preactivations->activations in $\nabla f_i\left(h_i\right(x){)}^T=\nabla f_{i+1}(h_{i+1}\left(x\right){)}^T{W}_i$, i.e., it should really say $$\nabla f_i\left(h_i\right(x){)}^T=\nabla f_{i+1}(h_{i+1}\left(x\right){)}^T{D}_{i+1}\left(x\right)W_i.$$ We now instead get $$W_i^T{W}_i\underset{\sim }{\propto }{W}_i^T\left(\sum _{x\in D}{D}_{i+1}\right(x\left)\nabla f_{i+1}\right(h_{i+1}\left(x\right)\left)\nabla f_{i+1}\right(h_{i+1}\left(x\right){)}^T{D}_{i+1}\left(x\right))W_i.$$ This should be the same as ${\sum }_{x\in D}{D}_{i+1}\left(x\right)\nabla f_{i+1}\left(h_{i+1}\right(x\left)\right)\nabla f_{i+1}\left(h_{i+1}\right(x\left){)}^T{D}_{i+1}\right(x)\underset{\sim }{\propto }I$ which, with $p_i$ denoting preactivations in layer $i$ and $f_{p,i}$ denoting the function from these preactivations to the output, is the same as $$\sum _{x\in D}\nabla f_{p,i+1}\left(p_{i+1}\right(x\left)\right)\nabla f_{p,i+1}\left(p_{i+1}\right(x){)}^T\underset{\sim }{\propto }I.$$ This last thing also totally works with activation functions other than ReLU — one can get this directly from the Jacobian calculation. I made the ReLU assumption earlier because I thought for a bit that one can get something further in that case; I no longer think this, but I won’t go back and clean up the presentation atm.

  Anyway, a takeaway is that the Deep Neural Feature Ansatz is equivalent to the (imo cleaner) ansatz that the set of gradients of the output wrt the pre-activations of any layer is close to being a tight frame (in other words, the gradients are in isotropic position; in other words still, the data matrix of the gradients is a constant times a semi-orthogonal matrix). (Note that the closeness one immediately gets isn’t in $L^2$ to a tight frame, it’s just in the quantity defining the tightness of a frame, but I’d guess that if it matters, one can also conclude some kind of closeness in $L^2$ from this (related).) This seems like a nicer fundamental condition because (1) we’ve intuitively canceled terms and (2) it now looks like a generic-ish condition, looks less mysterious, though idk how to argue for this beyond some handwaving about genericness, about other stuff being independent, sth like that.

  proof of the tight frame claim from the previous condition: Note that $$\sum _{x\in D}\nabla f_{p,i+1}\left(p_{i+1}\right(x\left)\right)\nabla f_{p,i+1}\left(p_{i+1}\right(x){)}^T\underset{\sim }{\propto }I$$clearly implies that the $L^2$ mass in any direction is the same, but also the $L^2$ mass being the same in any direction implies the above (because then, letting the SVD of the matrix with these gradients in its columns be $U^{\prime }{\Sigma }^{\prime }{V}^{\prime T}$, the above is $U^{\prime }{\Sigma }^{\prime }{\Sigma }^{\prime T}{U}^{\prime T}={\sigma }^{2}I$, where we used the fact that $\Sigma =\sigma I$).

  Some questions

  • Can one come up with some similar ansatz identity for the left singular vectors of $W_i$? One point of tension/​interest here is that an ansatz identity for $W_i{W}_i^T$ would constrain the left singular vectors of $W_i$ together with its singular values, but the singular values are constrained already by the deep neural feature ansatz. So if there were another identity for $W_i{W}_i^T$ in terms of some gradients, we’d get a derived identity from equality between the singular values defined in terms of those gradients and the singular values defined in terms of the Deep Neural Feature Ansatz. Or actually, there probably won’t be an interesting identity here since given the cancellation above, it now feels like nothing about $W_i$ is really pinned down by ‘gradients independent of $W_i$’ by the DNFA? Of course, some $W_i$-dependence remains even in the $i+1$ gradients because the preactivations at which further gradients get evaluated are somewhat $W_i$-dependent, so I guess it’s not ruled out that the DNFA constrains something interesting about $W_i$? But anyway, all this seems to undermine the interestingness of the DNFA, as well as the chance of there being an interesting similar ansatz for the left singular vectors of $W_i$.

  • Can one heuristically motivate that the preactivation gradients above should indeed be close to being in isotropic position? Can one use this reduction to provide simpler proofs of some of the propositions in the paper which say that the DNFA is exactly true in certain very toy cases?

  • The authors claim that the DNFA is supposed to somehow elucidate feature learning (indeed, they claim it is a mechanism of feature learning?). I take ‘feature learning’ to mean something like which neuronal functions (from the input) are created or which functions are computed in a layer in some broader sense (maybe which things are made linearly readable?) or which directions in an activation space to amplify or maybe less precisely just the process of some internal functions (from the input to internal activations) being learned of something like that, which happens in finite networks apparently in contrast to infinitely wide networks or NTK models or something like that which I haven’t yet understood? I understand that their heuristic identity on the surface connects something about a weight matrix to something about gradients, but assuming I’ve not made some index-off-by-one error or something, it seems to probably not really be about that at all, since the weight matrix sorta cancels out — if it’s true for one $W_i$, it would maybe also be true with any other $W_i$ replacing it, so it doesn’t really pin down $W_i$? (This might turn out to be false if the isotropy of preactivation gradients is only true for a very particular choice of $W_i$.) But like, ignoring that counter, I guess their point is that the directions which get stretched most by the weight matrix in a layer are the directions along which it would be the best to move locally in that activation space to affect the output? (They don’t explain it this way though — maybe I’m ignorant of some other meaning having been attributed to $W_i^T{W}_i$ in previous literature or something.) But they say “Informally, this mechanism corresponds to the approach of progressively re-weighting features in proportion to the influence they have on the predictions.”. I guess maybe this is an appropriate description of the math if they are talking about reweighting in the purely linear sense, and they take features in the input layer to be scaleless objects or something? (Like, if we take features in the input activation space to each have some associated scale, then the right singular vector identity no longer says that most influential features get stretched the most.) I wish they were much more precise here, or if there isn’t a precise interesting philosophical thing to be deduced from their math, much more honest about that, much less PR-y.

  • So, in brief, instead of “informally, this mechanism corresponds to the approach of progressively re-weighting features in proportion to the influence they have on the predictions,” it seems to me that what the math warrants would be sth more like “The weight matrix reweights stuff; after reweighting, the activation space is roughly isotropic wrt affecting the prediction (ansatz); so, the stuff that got the highest weight has most effect on the prediction now.” I’m not that happy with this last statement either, but atm it seems much more appropriate than their claim.

  • I guess if I’m not confused about something major here (plausibly I am), one could probably add 1000 experiments (e.g. checking that the isotropic version of the ansatz indeed equally holds in a bunch of models) and write a paper responding to them. If you’re reading this and this seems interesting to you, feel free to do that — I’m also probably happy to talk to you about the paper.

  typos in the paper

  indexing error in the first displaymath in Sec 2: it probably should say $W_L$″, not $W_{L+1}$″

• Kaarel 4 Apr 2024 14:17 UTC

  4 points

  0

  on: kh’s Shortform

  A thread into which I’ll occasionally post notes on some ML(?) papers I’m reading

  I think the world would probably be much better if everyone made a bunch more of their notes public. I intend to occasionally copy some personal notes on ML(?) papers into this thread. While I hope that the notes which I’ll end up selecting for being posted here will be of interest to some people, and that people will sometimes comment with their thoughts on the same paper and on my thoughts (please do tell me how I’m wrong, etc.), I expect that the notes here will not be significantly more polished than typical notes I write for myself and my reasoning will be suboptimal; also, I expect most of these notes won’t really make sense unless you’re also familiar with the paper — the notes will typically be companions to the paper, not substitutes.

  I expect I’ll sometimes be meaner than some norm somewhere in these notes (in fact, I expect I’ll sometimes be simultaneously mean and wrong/​confused — exciting!), but I should just say to clarify that I think almost all ML papers/​posts/​notes are trash, so me being mean to a particular paper might not be evidence that I think it’s worse than some average. If anything, the papers I post notes about had something worth thinking/​writing about at all, which seems like a good thing! In particular, they probably contained at least one interesting idea!

  So, anyway: I’m warning you that the notes in this thread will be messy and not self-contained, and telling you that reading them might not be a good use of your time :)

A start­ing point for mak­ing sense of task struc­ture (in ma­chine learn­ing)

• Kaarel 22 Feb 2024 0:12 UTC

  8 points

  0

  on: Why does gen­er­al­iza­tion work?

  I’d be very interested in a concrete construction of a (mathematical) universe in which, in some reasonable sense that remains to be made precise, two ‘orthogonal pattern-universes’ (preferably each containing ‘agents’ or ‘sophisticated computational systems’) live on ‘the same fundamental substrate’. One of the many reasons I’m struggling to make this precise is that I want there to be some condition which meaningfully rules out trivial constructions in which the low-level specification of such a universe can be decomposed into a pair $(s_1,s_2)$ such that $s_1$ and $s_2$ are ‘independent’, everything in the first pattern-universe is a function only of $s_1$, and everything in the second pattern-universe is a function only of $s_2$. (Of course, I’d also be happy with an explanation why this is a bad question :).)

• Kaarel 8 Feb 2024 3:00 UTC

  18 points

  15

  on: More Hyphenation

  I find [the use of square brackets to show the merge structure of [a linguistic entity that might otherwise be confusing to parse]] delightful :)

Toward A Math­e­mat­i­cal Frame­work for Com­pu­ta­tion in Superposition

• Kaarel 3 Nov 2023 19:36 UTC

  5 points

  0

  in reply to: Eliezer Yudkowsky’s comment on: Does davi­dad’s up­load­ing moon­shot work?

  I’d be quite interested in elaboration on getting faster alignment researchers not being alignment-hard — it currently seems likely to me that a research community of unupgraded alignment researchers with a hundred years is capable of solving alignment (conditional on alignment being solvable). (And having faster general researchers, a goal that seems roughly equivalent, is surely alignment-hard (again, conditional on alignment being solvable), because we can then get the researchers to quickly do whatever it is that we could do — e.g., upgrading?)

Grokking, mem­o­riza­tion, and gen­er­al­iza­tion — a discussion

• Kaarel 25 Aug 2023 14:31 UTC

  2 points

  0

  in reply to: dr_s’s comment on: AI Reg­u­la­tion May Be More Im­por­tant Than AI Align­ment For Ex­is­ten­tial Safety

  I was just claiming that your description of pivotal acts /​ of people that support pivotal acts was incorrect in a way that people that think pivotal acts are worth considering would consider very significant and in a way that significantly reduces the power of your argument as applying to what people mean by pivotal acts — I don’t see anything in your comment as a response to that claim. I would like it to be a separate discussion whether pivotal acts are a good idea with this in mind.

  Now, in this separate discussion: I agree that executing a pivotal act with just a narrow, safe, superintelligence is a difficult problem. That said, all paths to a state of safety from AGI that I can think of seem to contain difficult steps, so I think a more fine-grained analysis of the difficulty of various steps would be needed. I broadly agree with your description of the political character of pivotal acts, but I disagree with what you claim about associated race dynamics — it seems plausible to me that if pivotal acts became the main paradigm, then we’d have a world in which a majority of relevant people are willing to cooperate /​ do not want to race that much against others in the majority, and it’d mostly be a race between this group and e/​acc types. I would also add, though, that the kinds of governance solutions/​mechanisms I can think of that are sufficient to (for instance) make it impossible to perform distributed training runs on consumer devices also seem quite authoritarian.

• Kaarel 25 Aug 2023 0:34 UTC

  3 points

  0

  in reply to: dr_s’s comment on: AI Reg­u­la­tion May Be More Im­por­tant Than AI Align­ment For Ex­is­ten­tial Safety

  In this comment, I will be assuming that you intended to talk of “pivotal acts” in the standard (distribution of) sense(s) people use the term — if your comment is better described as using a different definition of “pivotal act”, including when “pivotal act” is used by the people in the dialogue you present, then my present comment applies less.

  I think that this is a significant mischaracterization of what most (? or definitely at least a substantial fraction of) pivotal activists mean by “pivotal act” (in particular, I think this is a significant mischaracterization of what Yudkowsky has in mind). (I think the original post also uses the term “pivotal act” in a somewhat non-standard way in a similar direction, but to a much lesser degree.) Specifically, I think it is false that the primary kinds of plans this fraction of people have in mind when talking about pivotal acts involve creating a superintelligent nigh-omnipotent infallible FOOMed properly aligned ASI. Instead, the kind of person I have in mind is very interested in coming up with pivotal acts that do not use a general superintelligence, often looking for pivotal acts that use a narrow superintelligence (for instance, a narrow nanoengineer) (though this is also often considered very difficult by such people (which is one of the reasons they’re often so doomy)). See, for instance, the discussion of pivotal acts in https://​​www.lesswrong.com/​​posts/​​7im8at9PmhbT4JHsW/​​ngo-and-yudkowsky-on-alignment-difficulty.

• Kaarel 27 Jun 2023 20:40 UTC

  LW: 4 AF: 2

  0

  AF

  on: Poly­se­man­tic­ity and Ca­pac­ity in Neu­ral Networks

  A few notes/​questions about things that seem like errors in the paper (or maybe I’m confused — anyway, none of this invalidates any conclusions of the paper, but if I’m right or at least justifiably confused, then these do probably significantly hinder reading the paper; I’m partly posting this comment to possibly prevent some readers in the future from wasting a lot of time on the same issues):

  
  1) The formula for $\tilde{y}$ here seems incorrect:
  

  

  
  
  This is because W_i is a feature corresponding to the i’th coordinate of x (this is not evident from the screenshot, but it is evident from the rest of the paper), so surely what shows up in this formula should not be W_i, but instead the i’th row of the matrix which has columns W_i (this matrix is called W later). (If one believes that W_i is a feature, then one can see this is wrong already from the dimensions in the dot product $W_i\cdot x$ not matching.)
  

  
  2) Even though you say in the text at the beginning of Section 3 that the input features are independent, the first sentence below made me make a pragmatic inference that you are not assuming that the coordinates are independent for this particular claim about how the loss simplifies (in part because if you were assuming independence, you could replace the covariance claim with a weaker variance claim, since the 0 covariance part is implied by independence):
  

  

  However, I think you do use the fact that the input features are independent in the proof of the claim (at least you say “because the x’s are independent”):

  

  Additionally, if you are in fact just using independence in the argument here and I’m not missing something, then I think that instead of saying you are using the moment-cumulants formula here, it would be much much better to say that independence implies that any term with an unmatched index is $0$. If you mean the moment-cumulants formula here https://​​en.wikipedia.org/​​wiki/​​Cumulant#Joint_cumulants , then (while I understand how to derive every equation of your argument in case the inputs are independent), I’m currently confused about how that’s helpful at all, because one then still needs to analyze which terms of each cumulant are 0 (and how the various terms cancel for various choices of the matching pattern of indices), and this seems strictly more complicated than problem before translating to cumulants, unless I’m missing something obvious.
  
  3) I’m pretty sure this should say x_i^2 instead of x_i x_j, and as far as I can tell the LHS has nothing to do with the RHS:

  

  (I think it should instead say sth like that the loss term is proportional to the squared difference between the true and predictor covariance.)

Crys­tal Heal­ing — or the Ori­gins of Ex­pected Utility Maximizers

• Kaarel 8 Jun 2023 23:47 UTC

  1 point

  −2

  on: Ques­tion for Pre­dic­tion Mar­ket peo­ple: where is the money sup­posed to come from?

  At least ignoring legislation, an exchange could offer a contract with the same return as S&P 500 (for the aggregate of a pair of traders entering a Kalshi-style event contract); mechanistically, this index-tracking could be supported by just using the money put into a prediction market to buy VOO and selling when the market settles. (I think.)

• Kaarel 5 Mar 2023 10:37 UTC

  4 points

  0

  on: kh’s Shortform

  An attempt at a specification of virtue ethics

  I will be appropriating terminology from the Waluigi post. I hereby put forward the hypothesis that virtue ethics endorses an action iff it is what the better one of Luigi and Waluigi would do, where Luigi and Waluigi are the ones given by the posterior semiotic measure in the given situation, and “better” is defined according to what some [possibly vaguely specified] consequentialist theory thinks about the long-term expected effects of this particular Luigi vs the long-term effects of this particular Waluigi. One intuition here is that a vague specification could be more fine if we are not optimizing for it very hard, instead just obtaining a small amount of information from it per decision.

  In this sense, virtue ethics literally equals continuously choosing actions as if coming from a good character. Furthermore, considering the new posterior semiotic measure after a decision, in this sense, virtue ethics is about cultivating a virtuous character in oneself. Virtue ethics is about rising to the occasion (i.e. the situation, the context). It’s about constantly choosing the Luigi in oneself over the Waluigi in oneself (or maybe the Waluigi over the Luigi if we define “Luigi” as the more likely of the two and one has previously acted badly in similar cases or if the posterior semiotic measure is otherwise malign). I currently find this very funny, and, if even approximately correct, also quite cool.

  Here are some issues/​considerations/​questions that I intend to think more about:

  1. What’s a situation? For instance, does it encompass the agent’s entire life history, or are we to make it more local?

  2. Are we to use the agent’s own semiotic measure, or some objective semiotic measure?

  3. This grounds virtue ethics in consequentialism. Can we get rid of that? Even if not, I think this might be useful for designing safe agents though.

  4. Does this collapse into cultivating a vanilla consequentialist over many choices? Can we think of examples of prompting regimes such that collapse does not occur? The vague motivating hope I have here is that in the trolley problem case with the massive man, the Waluigi pushing the man is a corrupt psycho, and not a conflicted utilitarian.

  5. Even if this doesn’t collapse into consequentialism from these kinds of decisions, I’m worried about it being stable under reflection, I guess because I’m worried about the likelihood of virtue ethics being part of an agent in reflective equilibrium. It would be sad if the only way to make this work would be to only ever give high semiotic measure to agents that don’t reflect much on values.

  6. Wait, how exactly do we get Luigi and Waluigi from the posterior semiotic measure? Can we just replace this with picking the best character from the most probable few options according to the semiotic measure? Wait, is this just quantilization but funnier? I think there might be some crucial differences. And regardless, it’s interesting if virtue ethics turns out to be quantilization-but-funnier.

  7. More generally, has all this been said already?

  8. Is there a nice restatement of this in shard theory language?

Search­ing for a model’s con­cepts by their shape – a the­o­ret­i­cal framework

• Kaarel 10 Feb 2023 2:46 UTC

  3 points

  0

  on: kh’s Shortform

  A small observation about the AI arms race in conditions of good infosec and collaboration

  Suppose we are in a world where most top AI capabilities organizations are refraining from publishing their work (this could be the case because of safety concerns, or because of profit motives) + have strong infosec which prevents them from leaking insights about capabilities in other ways. In this world, it seems sort of plausible that the union of the capabilities insights of people at top labs would allow one to train significantly more capable models than the insights possessed by any single lab alone would allow one to train. In such a world, if the labs decide to cooperate once AGI is nigh, this could lead to a significantly faster increase in capabilities than one might have expected otherwise.

  (I doubt this is a novel thought. I did not perform an extensive search of the AI strategy/​governance literature before writing this.)

[RFC] Pos­si­ble ways to ex­pand on “Dis­cov­er­ing La­tent Knowl­edge in Lan­guage Models Without Su­per­vi­sion”.

• Kaarel 20 Jan 2023 4:32 UTC

  2 points

  1

  on: How “Dis­cov­er­ing La­tent Knowl­edge in Lan­guage Models Without Su­per­vi­sion” Fits Into a Broader Align­ment Scheme

  First, suppose GPT-n literally just has a “what a human would say” feature and a “what do I [as GPT-n] actually believe” feature, and those are the only two consistently useful truth-like features that it represents, and that using our method we can find both of them. This means we literally only need one more bit of information to identify the model’s beliefs.

  One difference between “what a human would say” and “what GPT-n believes” is that humans will know less than GPT-n. In particular, there should be hard inputs that only a superhuman model can evaluate; on these inputs, the “what a human would say” feature should result in an “I don’t know” answer (approximately ⁵⁰⁄₅₀ between “True” and “False”), while the “what GPT-n believes” feature should result in a confident “True” or “False” answer.^[2] This would allow us to identify the model’s beliefs from among these two options.

  
  For $n$ such that GPT-$n$ is superhuman, I think one could alternatively differentiate between these two options by checking which is more consistent under implications, by which I mean that whenever the representation says that the propositions $P$ and $P\to Q$ are true, it should also say that $Q$ is true. (Here, for a language model, $P$ and $Q$ could be ~whatever assertions written in natural language.) Or more generally, in addition to modus ponens, also construct new propositions with ANDs and ORs, and check against all the inference rules of zeroth-order logic, or do this for first-order logic or whatever. (Alternatively, we can also write down versions of these constraints that apply to probabilities.) Assuming [more intelligent ⇒ more consistent] (w.r.t. the same set of propositions), for a superhuman model, the model’s beliefs would probably be the more consistent feature. (Of course, one could also just add these additional consistency constraints directly into the loss in CCS instead of doing a second deductive step.)

  I think this might even be helpful for differentiating the model’s beliefs from what it models some other clever AI as believing or what it thinks would be true in some fake counterfactual world, because presumably it makes sense to devote less of one’s computation to ironing out incoherence in these counterfactuals – for humans, it certainly seems computationally much easier to consistently tell the truth than to consistently talk about what would be the case in some counterfactual of similar complexity to reality (e.g. to lie).

  Hmm, after writing the above, now that I think more of it, I guess it seems plausible that the feature most consistent under negations is already more likely to be the model’s true beliefs, for the same reasons as what’s given in the above paragraph. I guess testing modus ponens (and other inference rules) seems much stronger though, and in any case that could be useful for constraining the search.

  (There are a bunch of people that should be thanked for contributing to the above thoughts in discussions, but I’ll hopefully have a post up in a few days where I do that – I’ll try to remember to edit this comment with a link to the post when it’s up.)

  What links here?

  • [RFC] Pos­si­ble ways to ex­pand on “Dis­cov­er­ing La­tent Knowl­edge in Lan­guage Models Without Su­per­vi­sion”. by gekaklam (25 Jan 2023 19:03 UTC; 47 points)