latent reasoning
In plain wordsGetting an AI to do its step-by-step thinking inside its internal numbers rather than writing it out in words — potentially faster and more flexible, but much harder for people to inspect.
Multi-step reasoning carried out in a model's internal activationsactivation, n.: The numbers a neural network computes inside itself as it processes an input — a reading of which of its artificial neurons are firing, and how strongly, at that moment. rather than in generated text tokens. It includes methods that feed the model's last hidden state back in as the next input ('continuous thought') instead of decoding a word, recurrent-depth architectures that iterate a block of layers more times on harder inputs, and training that internalizes an explicit chain of thoughtchain of thought, n.: Getting an AI to write out its reasoning step by step before giving an answer, the way a student shows their working. This often makes answers to math and logic problems more accurate. until it is no longer written out; the term also names the implicit multi-hop computation that interpretability researchers find inside ordinary forward passes. Proponents argue that continuous representations can hold several candidate paths at once and avoid spending tokens on language; safety researchers warn that moving reasoning out of readable text would erode chain-of-thought monitoringchain-of-thought monitoring, n.: Having a program — often another AI — read the step-by-step reasoning an AI writes before it acts, to catch signs that it intends to cheat, deceive or cause harm. and monitorabilitymonitorability, n.: How easy it is to watch an AI's reasoning and spot when it is planning something bad — a property that could be lost if future AIs stop thinking in readable language..
If labs switch to latent reasoning, we lose the cheapest window we have into what models are planning.
Origin
Shibo Hao and colleagues at Meta introduced Coconut ('chain of continuous thought') in December 2024, and Jonas Geiping and colleagues' recurrent-depth model (February 2025) scaled test-time computation in latent space. A July 2025 position paper by researchers from several laboratories urged developers to preserve monitorable text reasoning.
Related
- chain of thoughtchain of thought, n.: Getting an AI to write out its reasoning step by step before giving an answer, the way a student shows their working. This often makes answers to math and logic problems more accurate.
- reasoning tracereasoning trace, n.: The written-out 'thinking' an AI reasoning model produces before its final answer — notes to itself full of attempts, checks and corrections, which users may or may not get to see.
- monitorabilitymonitorability, n.: How easy it is to watch an AI's reasoning and spot when it is planning something bad — a property that could be lost if future AIs stop thinking in readable language.
- chain-of-thought monitoringchain-of-thought monitoring, n.: Having a program — often another AI — read the step-by-step reasoning an AI writes before it acts, to catch signs that it intends to cheat, deceive or cause harm.
- test-time computetest-time compute, n.: The computing effort an AI model spends while answering a question, as opposed to while being trained. Letting a model think longer or try many answers can make it markedly smarter.
- residual streamresidual stream, n.: The running internal state that flows through a transformer model from layer to layer; each layer reads from it and adds its own contribution, like editors annotating a shared document.
- chain of thoughtchain of thought, n.: Getting an AI to write out its reasoning step by step before giving an answer, the way a student shows their working. This often makes answers to math and logic problems more accurate.
Sources
- Hao, S., Sukhbaatar, S., Su, D., Li, X., Hu, Z., Weston, J. & Tian, Y. (2024). Training Large Language Models to Reason in a Continuous Latent Space. arXiv:2412.06769. arxiv.org/abs/2412.06769
Mentioned in
reasoning tracereasoning trace, n.: The written-out 'thinking' an AI reasoning model produces before its final answer — notes to itself full of attempts, checks and corrections, which users may or may not get to see., hierarchical reasoning modelhierarchical reasoning model, n.: A small, brain-inspired AI model that solves puzzles with two linked parts — a slow planner and a fast worker — instead of writing out its reasoning in words., tiny recursive modeltiny recursive model, n.: A very small AI model that solves hard puzzles by going over its own answer again and again, improving it each time, instead of relying on sheer size.