Antithetical Labs
antithetical-labs.com
Actively hiring
AI Safety & Alignment
AI Evaluation & Observability
Follow
Overview
News
Technologies
Salaries
Products
People
Growth
Offices
Financials
Overview
Antithetical Labs is a one-person lab. We measure how AI models actually behave, publish the numbers, and build with what the data says.
News
The judge liked it better without the citations
I deleted every citation from a research memo then asked an LLM judge to score both the original and the stripped versions. The stripped one scored higher. Removing all evidence citations across ten memos only moved the judge's score by +0.17. When the me
Read more
Report
Convergence you chose
Last time I measured what four models build when the brief says nothing about design. They converge, the palette follows your product category, and telling them to "be distinctive" strips the gloss without moving the design anywhere. So: what does move it
Read more
Report
The most likely page
I gotta confess: I have no design training and a pretty utilitarian taste, so when I need a UI I just ask a model and then squint at it until I am satisfied. Lately I've been squinting more. The output has a sameness to it that I can spot and couldn't des
Read more
Report