[2503.18866] Reasoning to Learn from Latent Thoughts

Abstract:Compute scaling for language model (LM) pretraining has outpaced the growth of human-written texts, leading to concerns that data will become the bottleneck to LM scaling. To continue scaling pretraining in this data-constrained regime, we propose that explicitly modeling and inferring the latent thoughts that underlie the text generation process can significantly improve pretraining data efficiency. Intuitively, our approach views web text as the compressed final outcome of a verbose human thought process and that the latent thoughts contain important contextual knowledge and reasoning steps that are critical to data-efficient learning. We empirically demonstrate the effectiveness of our approach through data-constrained continued pretraining for math. We first show that synthetic data approaches to inferring latent thoughts significantly improve data efficiency, outperforming training on the same amount of raw data (5.7\% $\rightarrow$ 25.4\% on MATH). Furthermore, we demonstrate latent thought inference without a strong teacher, where an LM bootstraps its own performance by using an EM algorithm to iteratively improve the capability of the trained LM and the quality of thought-augmented pretraining data. We show that a 1B LM can bootstrap its performance across at least three iterations and significantly outperform baselines trained on raw data, with increasing gains from additional inference compute when performing the E-step. The gains from inference scaling and EM iterations suggest new opportunities for scaling data-constrained pretraining.

Reasoning to Learn from Latent Thoughts Yangjun Ruan1,2,3 , Neil Band1 , Chris J. Maddison2,3† , Tatsunori Hashimoto1† 1 Stanford University 2 University of Toronto 3 Vector Institute {yjruan,cmaddis}@cs.toronto.edu, {nband,thashim}@stanford.edu Abstract Compute scaling for language model (LM) pretraining has outpaced the…

related reading

• As Rocks May Think | Eric Jangevjang.com
• Training Large Language Models to Reason in a Continuous Latent Spacearxiv.org
• [2509.14786] Pre-training under infinite computearxiv.org
• [2412.06769] Training Large Language Models to Reason in a Continuous Latent Spacearxiv.org
• The Extreme Inefficiency of RL for Frontier Models - Toby Ordtobyord.com
• [2201.11903] Chain of Thought Prompting Elicits Reasoning in Large Language Modelsarxiv.org
• Unlocking the Working Memory of Large Language Models for Latent Reasoningarxiv.org
• [2502.19402] General Reasoning Requires Learning to Reason from the Get-goar5iv.labs.arxiv.org
• [2005.14165] Language Models are Few-Shot Learnersarxiv.org
• MAI-Thinking-1: Building a Hill-Climbing Machinemicrosoft.ai
• Synthetic Pretraining | Vintage Datavintagedata.org
• Will scaling work? - by Dwarkesh Patel - Dwarkesh Podcastdwarkesh.com