Semilattice

Blog

The latest from our quest to advance decision-making.

Get new posts in your inbox
Simulate users in any language
User models Product

Simulate users in any language

We can now build user models of people who speak languages other than English. They're built from real human data in that language and validated to the same state-of-the-art accuracy as English models.

Joseph Wright Joseph Wright
Anacreon: state-of-the-art accuracy in simulating individuals
Research Company User models

Anacreon: state-of-the-art accuracy in simulating individuals

Anacreon predicts how individual people answer new questions, not just population averages, preserving diverse and minority group opinions. It achieves a state-of-the-art ordinal alignment of 0.775, the individual-level accuracy measure on which the field has converged.

Pranav Dahiya Pranav Dahiya
How MONEYME settled a positioning question that had been open for months
Company User models

How MONEYME settled a positioning question that had been open for months

MONEYME launches credit products and changes its messaging more often than research with real people can keep up with. Its teams now have a model of 450 everyday Australians, each one its own fine-tuned LLM, that they can consult in minutes.

Fabian Teichmueller Fabian Teichmueller
What we found when we audited 1,000 predictions
User models Product

What we found when we audited 1,000 predictions

Sometimes a Semilattice prediction lands with one answer at 99.8%, or two answers stacked at 99.9% combined. We decided to run an audit to diagnose what was happening.

Joseph Wright Joseph Wright