Scaling Latent Reasoning via Looped Language Models

If you squint your eyes it's a fixed iteration ODE solver. I'd love to see a generalization on this and the Universal Transformer metioned re-envisioned as flow-matching/optimal transport models.

Does the training process ensure that all the intermediate steps remain interepretable, even on larger models? Not that we end up with some alien gibberish in all but the final step.

so it's:

output = layers(layers(layers(layers(input))))

instead of the classical:

output = layer4(layer3(layer2(layer1(input))))