Lovre

Contact at [my name].pesut@gmail.com

That seems like a great idea, and induction heads do seem highly relevant!

What you describe is actually one of the key reasons why I'm so excited about this whole approach. I've seen many interesting metalearning tasks, and they mostly just like work or not work, or they fail sometimes, and you can try to study their failures to perhaps glean some insight into the underlying algorithm -- but...they just don't have (m)any nontrivial "degrees of freedom" in which you can vary them. The class of numerical models, on the other hand, has a substantial amount of nontrivial ways in which you can vary your input -- and even more so, you can vary it not just discretely, but also ~continuously.

That makes me really optimistic about the possibility of which you hint, of reverse engineering whatever algorithm the model is running underneath, and then using interpretability tools to verify/falsify those findings. Conversely, interpretability tools could be used to make predictions about the algorithm, which can then be checked. Hence one can imagine a quite meaningful feedback loop between experimentation and interpretability!

Since I transformed the Iris dataset with a pretty "random" transformation (i.e. not chosen because it was particularly nice in some way), I didn't check for its regeneration -- since my feature vectors were very different to original Iris's, and it seemed exceedingly unlikely that feature vectors were saved anywhere on the internet with that particular transformation.

But I got curious now, so I performed some experiments.

So these are the first 7 transformed feature vectors (in one of the random samplings of the dataset). Among all the generated output (I looked at >200 vectors), it never once output a vector which was identical to any of the latter ones, and also... in general the stuff it was generating did not look like it was drawing on any knowledge of the remaining vectors in the dataset. (E.g. it generated a lot that were off-distribution.)

I also tried

Where I cherrypicked the "class 2" so that the first coordinate is lower than usual for that class; and the generated stuff always had the first coordinate very off-distribution from the rest of the class 2, as one would expect if the model was meta-learning from the vectors it sees, rather than "remembering" something.

This last experiment might seem a little contrived, but bit of a problem with this kind of testing is that if you supply enough stuff in-context, the model (plausibly) meta-learns the distribution and then can generate pretty similar-looking vectors. So, yeah, to really get to the bottom of this, to become 'certain' as it were, I think one would have to go in deeper than just looking at what the model generates.

(Or maybe there are some ways around that problem which I did not think of. Suggestions appreciated!)

To recheck things again -- since I'm as worried about leakage as anyone -- I retested Iris, this time transforming each coordinate with its own randomly-chosen affine transformation: (x1,x2,x3,x4)↦(11x1+3,7x2+18,9x3+5,22x4+12)

And the results are basically identical to those with just one affine transformation for all coordinates.

I'm glad that you asked about InstructGPT since I was pretty curious about that too, was waiting for an excuse to test it. So here are the synthetic binary results for (Davinci-)InstructGPT, compared with the original Davinci from the post:

ModelScen. 1Scen. 2Scen. 3Scen. 4Scen. 5Scen. 6Scen. 7Scen. 8Scen. 9Vanilla GPT-367.78%76.67%77.78%82.22%95.56%77.78%70.0%72.22%63.33%Instruct GPT-375.56%72.22%71.11%86.67%95.56%77.78%84.44%73.33%73.33%