Somabets Information

Every decision is a bet. Make a good one. SomaBets is the AI-data brand of Sensie, LLC — a verification layer for the human feedback that trains AI models. The problem: annotator inattention and LLM-assisted completion are invisible at the label level. A disengaged rater produces plausible labels that pass QA and poison the reward model anyway. A rater pasting tasks into a frontier model produces output that is, by construction, indistinguishable from engaged human judgment. Every QA system inspects the label. Nothing inspects the human. SomaBets inspects the human. A 3-second smartphone motion read — involuntary micro-motion, no wearable, classified server-side — detects whether a person's nervous-system state is congruent with the judgment they're stating. ALIGNED or UNALIGNED. The signal is involuntary, so it isn't under conscious control. Not just proof that a human did the task: proof the human was behind the answer. The product is a routing layer, not a polygraph. Our pre-screen scores per-annotator signal reliability before labels enter your pipeline, so you pay for verified reads from high-signal humans. The evidence: 9 PhD-led research trials, 18,000+ sessions, 83.6% accuracy where the read locks in, two granted US utility patents. Built on 13 years of research with scientists affiliated with Harvard Medical School and Oxford. We run fixed-scope paid pilots on our managed panel with pre-registered success criteria — you validate against your own gold set. somabets.com · a Sensie, LLC brand · St. Petersburg, FL