Trismik, a Cambridge spin-out, secured £2.2 million in Pre-Seed funding, led by Twinpath Ventures.
The capital will launch its adaptive AI evaluation platform. Traditional benchmarks for large language models often saturate. Trismik addresses this problem with a science-backed approach to AI evaluation.
The company applies psychometric methods, including Item Response Theory and Computerised Adaptive Testing. This technique, borrowed from human IQ testing, adapts evaluation difficulty in real-time. It delivers precise accuracy rankings with fewer test items.
Trismik's platform significantly cuts evaluation costs, potentially by up to 95 percent. It also boosts measurement precision. Professor Nigel Collier, Trismik's Chief Scientific Officer, pioneered this scientific approach. Rebekka Mikkola serves as CEO, with Marco Basaldella as CTO.
The platform supports classical and adaptive testing for AI evaluation across various datasets. These include factuality, alignment, reasoning, and safety. Trismik envisions its platform evolving into a broader environment for machine learning experimentation. This will incorporate fine-tuning and compliance tracking.
