AI CoT Reasoning Is Often Unfaithful

from Don't Worry About the Vase [alt+shift+b] in AI

A new Anthropic paper reports that reasoning model chain of thought (CoT) is often unfaithful. They test on Claude Sonnet 3.7 and r1, I’d love to see someone try this on o3 as well.

2 months ago

Remove from reading list Add to reading list [alt+a] Read now [→]

Improve your reading experience

Logged in users get linked directly to articles resulting in a better reading experience. Please login for free, it takes less than 1 minute.

More from Don't Worry About the Vase

Cheaters Gonna Cheat Cheat Cheat Cheat Cheat

Cheaters.

a month ago • 16 votes

AI #115: The Evil Applications Division

It can be bleak out there, but the candor is very helpful, and you occasionally get a win.

a month ago • 16 votes

OpenAI Claims Nonprofit Will Retain Nominal Control

Your voice has been heard.

a month ago • 18 votes

Zuckerberg's Dystopian AI Vision

You think it’s bad now?

a month ago • 15 votes

GPT-4o Sycophancy Post Mortem

Last week I covered that GPT-4o was briefly an (even more than usually) absurd sycophant, and how OpenAI responded to that.

a month ago • 19 votes

More in AI

Notes on Japan

19 hours ago • 4 votes

Weekly ML for SWEs #12: The easiest way to keep up with ML and AI research papers

An AI reading list curated to make you a better engineer: 6-17-25

18 hours ago • 2 votes

AI Roundup 122: Economies of Scale

June 13, 2025.

5 days ago • 4 votes

How to afford Machine Learning for Software Engineers

How I'm planning to make ML for SWEs more helpful while remaining affordable

6 days ago • 6 votes

The State of AI Engineering (2025)

Big ideas from the 2025 World's Fair.

a week ago • 7 votes

New here?