We study reasoning in language models and build systems grounded in that understanding.
mnemic labs is an independent research lab that publishes research and develops production-ready systems for partner organizations.
Our position
A model answers with the same confidence whether it knows or guesses. Our work is making reasoning checkable: show the steps, test each one, find where it breaks.
Half of the lab is research, published in the open. The other half is a small number of engagements each year, building and deploying systems inside an organization's own environment. Each side keeps the other honest.
Selected research
Measuring faithfulness in chain-of-thought reasoning
A method for testing whether a model's stated reasoning reflects the computation behind its answer, applied to four open models.
Self-correction helps only when the model can verify
A controlled study of when asking a model to check its own work improves accuracy, and when it does not.
What production deployments taught us about evaluation
Notes from running question answering systems inside real organizations: what held up, what did not, and what we now test before handover.
Principles
Small teams, full context
One person owns a question from start to finish.
Experiments settle arguments
When we disagree, we run the experiment.
Negative results are results
Failed replications get written up like successes.
Tools before demos
We ship software others can use, not videos.
Claims fit the evidence
We say what the data supports, and stop there.
Careers
There are no open roles right now. When one opens, it will be on the careers page.
Contact