Bridging the Gap Between Latent and Explicit Reasoning with Looped Transformers
LOTUS repeatedly refines blocks of continuous reasoning states in parallel, training each position to predict a token from a worked solution. It generates answers from those states and studies accuracy, latency and how much of the reasoning can be read back.
First public: · Added to this list: October 23, 2026
Explore this approach
Internalized reasoning — Learn from worked solutions so that some or all of the reasoning can later happen without writing those steps. Training may use a separate teacher model or progressively remove written steps from the model being trained.
Continuous-state reasoning — Use vectors—lists of numbers—to carry intermediate reasoning from one step to the next, instead of selecting a text token at every step. Some methods use the model's hidden states; others combine several candidate token representations.
Recurrent reasoning — Repeat parts of the model or keep intermediate information in memory while working through a problem. Some methods also learn how many passes to run.
Alternative generation methods — Change how intermediate steps are generated, revised, or searched—for example, by refining several steps in parallel. Some produce written chains of thought.