Theoretical Perspective on Continuous Chain of Thoughts in Reasoning
By
danielmorozoff
11mo ago· 2 min readenInsight
75/100
Toasty
Bagelometer↗
A bagel you'd recommend to a friend without hedging.
Score75TypeanalysisSentimentneutral
Summary
Large Language Models (LLMs) have shown impressive performance in reasoning tasks using chain-of-thoughts (CoTs) techniques. This article explores the theoretical perspective on continuous CoTs and their superiority over discrete tokens in solving reasoning problems like directed graph reachability.
Key quotes
· 3 pulledWhile existing theoretical works demonstrate that CoTs with discrete tokens boost the capability of LLMs, recent work on continuous CoTs lacks a theoretical understanding of why it outperforms discrete counterparts in various reasoning tasks.
In our construction, each continuous thought vector is a superposition state that encodes multiple search frontiers simultaneously (i.e., parallel breadth-first search (BFS)).
Encoding of multiple search frontiers as a superposition state automatically emerges in training continuous CoTs, without explicit supervision to guide the model to explore multiple paths simultaneously.
Large Language Models (LLMs) have demonstrated remarkable performance in many applications, including challenging reasoning problems via chain-of-thoughts (CoTs) techniques that generate ``thinking tokens'' before answering the questions. While existing t
You might also wanna read
Researchers Develop Method to Predict Real-Time Progress in Reasoning Language Models
This research paper investigates whether real-time progress prediction is feasible for reasoning language models that use long latent chains
arxiv.org·4d ago
Study Reveals How RL and SFT Differently Teach Transformers Chain-of-Thought Reasoning on Sparse Boolean Functions
This research paper analyzes how transformers learn Chain-of-Thought (CoT) reasoning capabilities through Reinforcement Learning (RL) with p
arxiv.org·3d ago