Connor Dilgren

CS MS student @umdcs

[Connor Dilgren @ConnorDilgren]

@frisbeemortel @mariusmosbach Wow nice work Michael! Totally agree with your conclusion that fine-tuned models don't seem to use latent reasoning (let alone superposition!). And it's great you found that from-scratch models can learn it. Happy to chat!

[Connor Dilgren @ConnorDilgren]

Excited to announce my first preprint in LM interpretability!

Latent reasoning models are not monitorable by default, since they don't reason in human-readable, natural language text. But can we make progress in understanding their intermediate reasoning steps using mech interp?

[Connor Dilgren @ConnorDilgren]

Thanks to @sarahwiegreffe for advising this project!

Read the preprint here: https://t.co/1b8Cu1TUjj

[Connor Dilgren @ConnorDilgren]

Overall, these results are somewhat encouraging for latent reasoning model interpretability. But I suspect models with weaker natural language priors, such as those trained to do latent reasoning during pretraining or through RL, will be much less interpretable.