BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Implicit Chain-of-Thought: Internalizing Reasoning in Language Mod
 els  - Yuntian Deng (University of Waterloo)
DTSTART:20251016T150000Z
DTEND:20251016T160000Z
UID:TALK237769@talks.cam.ac.uk
CONTACT:Lucas Resck
DESCRIPTION:Abstract:\nWhen leveraging language models for reasoning tasks
 \, generating explicit chain-of-thought (CoT) steps is often crucial for h
 igh accuracy. In this work\, drawing inspiration from how the human brain 
 transitions from explicit\, conscious\, deliberate reasoning (System 2) to
  implicit\, automatic\, intuitive thinking (System 1)\, we seek to interna
 lize explicit CoT reasoning within a model that directly produces the fina
 l answer\, which we define as the implicit CoT paradigm.\n\nTo realize imp
 licit CoT\, we found a simple yet effective method: starting with a model 
 trained for explicit CoT reasoning\, we gradually remove the intermediate 
 steps and finetune the model. This approach enables a finetuned GPT-2 Smal
 l model to solve 20-by-20 multiplication with up to 99.5% accuracy\, where
 as standard training cannot solve beyond 4-by-4 multiplication.\n\nYou can
  try our demo at https://huggingface.co/spaces/yuntian-deng/gpt2-multiplic
 ation\n---\n\nBio:\nYuntian Deng is an assistant professor at the Universi
 ty of Waterloo and a visiting professor at NVIDIA under Prof. Yejin Choi. 
 He was previously a postdoc at AI2\, also advised by Prof. Choi. He receiv
 ed his PhD from Harvard University under Prof. Alexander Rush and Prof. St
 uart Shieber. His recent works include NeuralOS\, Interactive Training\, W
 ildChat\, and Implicit Chain-of-Thought.\n
LOCATION:https://cam-ac-uk.zoom.us/j/97599459216?pwd=QTRsOWZCOXRTREVnbTJBd
 XVpOXFvdz09
END:VEVENT
END:VCALENDAR