Scaling up test-time compute with latent reasoning: A recurrent depth approach
1–10 of 45 posts
Re: Scaling up test-time compute with latent reasoning: A recurrent depth approach
#2Twitter thread about this by the author: https://x.com/jonasgeiping/status/1888985929727037514
Re: Scaling up test-time compute with latent reasoning: A recurrent depth approach
#3Interesting stuff. As the authors note, using latent reasoning seems to be a way to sink more compute into the model and get better performance without increasing the model size, good news for those on a steady diet of 'scale pills'
Re: Scaling up test-time compute with latent reasoning: A recurrent depth approach
#4Latent / embedding-space reasoning seems a step in the right direction, but building recurrence into the model while still relying on gradient descent (i.e. BPTT) to train it seems to create more of a problem (training inefficiency) than it solves, especially since they still end up externally specifying the number of recurrent iterations (r=4, 8, etc) for a given inference. Ideally having recurrence internal to the model would allow the model itself to decide how long to iterate for before outputting anything.
Re: Scaling up test-time compute with latent reasoning: A recurrent depth approach
#5Twitter thread about this by the author: https://x.com/jonasgeiping/status/1888985929727037514
If you don't have a twitter account and want to read the full thread: https://xcancel.com/jonasgeiping/status/1888985929727037514
Re: Scaling up test-time compute with latent reasoning: A recurrent depth approach
#6Latent / embedding-space reasoning seems a step in the right direction, but building recurrence into the model while still relying on gradient descent (i.e. BPTT) to train it seems to create more of a problem (training inefficiency) than it solves, especially since they still end up externally specifying the number of recurrent iterations (r=4, 8, etc) for a given inference. Ideally having recurrence internal to the…
> Latent / embedding-space reasoning seems a step in the right direction
Might be good for reasoning, but it's terrible for interpretation / AI-safety.
Re: Scaling up test-time compute with latent reasoning: A recurrent depth approach
#7One of the benefits of using thinking tokens compared to “thinking in a latent” space is that you can directly observe the quality of the CoT. In R1 they saw it was mixing languages and fixed it with cold start data.
It would be hard to SFT this because you can only SFT the final result not the latent space.
I also notice the authors only had compute for a single full training run. It’s impressive they saw such good results from that, but I wonder if they could get better results by incorporating recent efficiency improvements.
I would personally not use this architecture because 1) it adds a lot of hyperparameters which don’t have a strong theoretical grounding and 2) it’s not clearly better than simpler methods.
Re: Scaling up test-time compute with latent reasoning: A recurrent depth approach
#8Latent / embedding-space reasoning seems a step in the right direction, but building recurrence into the model while still relying on gradient descent (i.e. BPTT) to train it seems to create more of a problem (training inefficiency) than it solves, especially since they still end up externally specifying the number of recurrent iterations (r=4, 8, etc) for a given inference. Ideally having recurrence internal to the…
> seems a step in the right direction
I can’t see why. I can’t think of any problems where recurrent loops with latent streams would be preferable to tokens. And the downsides are obvious.
> externally specifying the number of recurrent iterations
Yeah this seems wrong to me. At least with RL training you saw that the length of the CoT decreased dramatically before climbing again, as the model became more proficient.
Re: Scaling up test-time compute with latent reasoning: A recurrent depth approach
#9My opinion is that opaque reasoning is a prerequisite for many of the worst possible AI outcomes.
We should make reasoning fully visible in the output space.
Re: Scaling up test-time compute with latent reasoning: A recurrent depth approach
#10Latent / embedding-space reasoning seems a step in the right direction, but building recurrence into the model while still relying on gradient descent (i.e. BPTT) to train it seems to create more of a problem (training inefficiency) than it solves, especially since they still end up externally specifying the number of recurrent iterations (r=4, 8, etc) for a given inference. Ideally having recurrence internal to the…
> seems a step in the right direction I can’t see why. I can’t think of any problems where recurrent loops with latent streams would be preferable to tokens. And the downsides are obvious. > externally specifying the number of recurrent iterations Yeah this seems wrong to me. At least with RL training you saw that the length of the CoT decreased dramatically before climbing again, as the model became more proficient.
Identifying scheming in the latent streams would be harder as you would have an extra layer of obfuscation between you and the model’s reasoning.