Saved in:
Bibliographic Details
Main Authors: McAleese, Nat, Pokorny, Rai Michael, Uribe, Juan Felipe Ceron, Nitishinskaya, Evgenia, Trebacz, Maja, Leike, Jan
Format: Preprint
Published: 2024
Subjects:
Online Access:https://arxiv.org/abs/2407.00215
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866914852345217024
author McAleese, Nat
Pokorny, Rai Michael
Uribe, Juan Felipe Ceron
Nitishinskaya, Evgenia
Trebacz, Maja
Leike, Jan
author_facet McAleese, Nat
Pokorny, Rai Michael
Uribe, Juan Felipe Ceron
Nitishinskaya, Evgenia
Trebacz, Maja
Leike, Jan
contents Reinforcement learning from human feedback (RLHF) is fundamentally limited by the capacity of humans to correctly evaluate model output. To improve human evaluation ability and overcome that limitation this work trains "critic" models that help humans to more accurately evaluate model-written code. These critics are themselves LLMs trained with RLHF to write natural language feedback highlighting problems in code from real-world assistant tasks. On code containing naturally occurring LLM errors model-written critiques are preferred over human critiques in 63% of cases, and human evaluation finds that models catch more bugs than human contractors paid for code review. We further confirm that our fine-tuned LLM critics can successfully identify hundreds of errors in ChatGPT training data rated as "flawless", even though the majority of those tasks are non-code tasks and thus out-of-distribution for the critic model. Critics can have limitations of their own, including hallucinated bugs that could mislead humans into making mistakes they might have otherwise avoided, but human-machine teams of critics and contractors catch similar numbers of bugs to LLM critics while hallucinating less than LLMs alone.
format Preprint
id arxiv_https___arxiv_org_abs_2407_00215
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle LLM Critics Help Catch LLM Bugs
McAleese, Nat
Pokorny, Rai Michael
Uribe, Juan Felipe Ceron
Nitishinskaya, Evgenia
Trebacz, Maja
Leike, Jan
Software Engineering
Machine Learning
Reinforcement learning from human feedback (RLHF) is fundamentally limited by the capacity of humans to correctly evaluate model output. To improve human evaluation ability and overcome that limitation this work trains "critic" models that help humans to more accurately evaluate model-written code. These critics are themselves LLMs trained with RLHF to write natural language feedback highlighting problems in code from real-world assistant tasks. On code containing naturally occurring LLM errors model-written critiques are preferred over human critiques in 63% of cases, and human evaluation finds that models catch more bugs than human contractors paid for code review. We further confirm that our fine-tuned LLM critics can successfully identify hundreds of errors in ChatGPT training data rated as "flawless", even though the majority of those tasks are non-code tasks and thus out-of-distribution for the critic model. Critics can have limitations of their own, including hallucinated bugs that could mislead humans into making mistakes they might have otherwise avoided, but human-machine teams of critics and contractors catch similar numbers of bugs to LLM critics while hallucinating less than LLMs alone.
title LLM Critics Help Catch LLM Bugs
topic Software Engineering
Machine Learning
url https://arxiv.org/abs/2407.00215