Interesting results to see that the original model is actually very faithful. This may well be a reverse-engineering exercise that shows the models have been trained for this (i.e. steering will move them away from the optimal mode). Relatedly, in alignment research, we often see the concept of 'corrigibility' pop up. I think the way you approach faithfulness to CoTs may well extend into this concept, allowing us to edit the model to be more able to acknowledge and fix its mistakes. There's an extension / version of this project that may well direct a model towards safety through feature-steered corrigibility. Exciting progress!

Great work! It’s very easy to understand what you were investigating and the results are also clearly presented. A little bit surprising to see that nearly 0 steering has the strongest accuracy, but the edge case results are insightful. I would be interested to see the effects of negative steering on pre instruct tuning vs post and seeing if negative steering is faithful if the models are allowed to generate their own CoT. Hope you continue this work as you already have interesting results.

This is an interesting result: the authors look at faithfulness in chain-of-thought reasoning and surprisngly find that single features can substantially alter faithfulness. The methodology is sensible and the experiments are carried out well and well-documented. The relation to safety is subtle but well-justified.

In this case the method used (adding mistakes) means that increased faithfulness leads to a decrease in performance. This is a sensible experimental methodology for understanding if features can control faithfulness - a cool result in its own right. In my anecdotal experience in LLM reasoning, incorrect answers typically arise from a lack of faithfulness to an otherwise correct chain of thought. It would be an interesting extension to the paper to see if the features identified in this paper lead to an increase in performance in natural chain of thought reasoning settings.

This team develops a reasonable experiment setup and executes it well. Their results point to an interesting possibility, that subtracting the "acknowledging mistakes" feature could lead to higher faithfulness.

With more time, a graph I would've liked to see is faithfulness by steering value.

I would also be interested in seeing the team explore whether allowing the model to continue the CoT will recover the faithfulness lost by this steering by acknowledging the reasoning error out loud.

Good work!