Update! It appears to be almost entirely slop. If I had to describe the main failure mode, I would say “The AIs tend to fail to focus on what kinds of things are valuable to me, instead bouncing off and producing similar-looking but useless results.”
For the hierarchical latents/hieararchical planning, it looks like the DeepSeeks reduced everything to gaussian mixtures which is no longer interesting.
For the PDG results on natural latents it’s unclear whether they did anything non-trivial at all. Having thought about it, it’s not obvious exactly what could be done for the latents work in PDGs, since I think John and David (in their original proofs) are secretly just using Bayes nets as PDGs without mentioning them. IIf we have latents and over and probability distributions and which agree with a “reference” probability distribution to within , then this is actually just already constructing a PDG with edges and which disagree. Whoops.
The Logical Inductors result might actually be non-trivial. It appears to state that no two logical inductors can accurately model each others prices for a given statement at time t i.e. for two logical inductors L and M, and a statement , the two inductors cannot, at time t, both have accurate predictions for each others’ price on . I will need to look into this further though, as this is not really my area of expertise and the proof is, as expected, difficult to interpret. I also think this might be pointless? The important thing is the actual bound on the errors. If the bound goes down provably over time that would be a pretty cool converse result.
I will try this again though, I plan to have them study the following questions:
In a PDG modelling active inference with a reference distribution over outcomes and an approximate distribution which differ by some amount, how much can our approximate distribution be incorrect, and how aggressive our optimization before doing RL-as-inference on causes a collapse in score in ?
Build on that, see whether the results hold if we have a hierarchical Q which factors as latents, likewise a hierarchical P, and whether this provides some evidence for what hierarchical planning actually looks like in the wild.
I’ll actually throw Fable and Astra at the problem this time, I think.
Update! It appears to be almost entirely slop. If I had to describe the main failure mode, I would say “The AIs tend to fail to focus on what kinds of things are valuable to me, instead bouncing off and producing similar-looking but useless results.”
For the hierarchical latents/hieararchical planning, it looks like the DeepSeeks reduced everything to gaussian mixtures which is no longer interesting.
For the PDG results on natural latents it’s unclear whether they did anything non-trivial at all. Having thought about it, it’s not obvious exactly what could be done for the latents work in PDGs, since I think John and David (in their original proofs) are secretly just using Bayes nets as PDGs without mentioning them. IIf we have latents
The Logical Inductors result might actually be non-trivial. It appears to state that no two logical inductors can accurately model each others prices for a given statement at time t i.e. for two logical inductors L and M, and a statement
I will try this again though, I plan to have them study the following questions:
In a PDG modelling active inference with a reference distribution over outcomes
Build on that, see whether the results hold if we have a hierarchical Q which factors as latents, likewise a hierarchical P, and whether this provides some evidence for what hierarchical planning actually looks like in the wild.
I’ll actually throw Fable and Astra at the problem this time, I think.