Blog
Research and engineering notes.
Research and engineering notes on uncertainty quantification, risk-aware machine learning, and trustworthy AI from the Themis AI team.
Trustworthy edge AI on the NXP i.MX 8M Plus: running a risk-aware vision model in full INT8 on the eIQ VX delegate
Read the demonstration
AI That Knows When Not to Act — Live at NXP Technology Days
Themis AI is at NXP Technology Days in Detroit, October 28–29, 2026. Two live demos on NXP silicon show what a model looks like when it knows its own limits.
Read
The Newest Decision Models Still Need a Decision Layer
We ran Liquid AI’s new d1.1 and d1-ultra and TypeSafe’s Jev 1.14 through the intent-routing tests from our last post. An open 3B model, finetuned and given the Themis Decision Layer, still came out ahead.
Read
Decision Models Need a Decision Layer
What we found running d1-3B on intent-routing benchmarks that include requests none of the options fit, and what the Themis Decision Layer adds on top.
Read
Before AI Acts
Why trustworthy decision layers belong alongside model-level safety.
Read
Quantifying Epistemic Uncertainty with the Laplace Approximation
Your trained model already knows what it doesn't know — it's hidden in the curvature of the loss landscape. The Laplace approximation extracts it, post hoc, with no retraining.
Read
Spotting the Unfamiliar: Out-of-Distribution Detection with Uncertainty
Models fail most dangerously on inputs they've never seen. Epistemic uncertainty flags those out-of-distribution cases in real time — so the system can escalate, abstain, or adapt.
Read
Managing Risk in Production AI Systems
Why production AI needs risk measurement, not just predictions — and how to turn a model's uncertainty into enforceable risk policies.
Read
NIST's Risk Management Framework and Themis AI
Examining the Risk Management Framework by NIST.
Read
Catching LLM Hallucinations with Uncertainty
LLMs fail without warning. A separate uncertainty estimate lets a model abstain, escalate, or defer instead of confidently making things up.
Read
Drug Discovery Cost Reduction with Uncertainty-Guided Predictions
Read
Uncertainty-Aware Human Intervention for Autonomous Vehicles
Read
Risk-Aware Selective Answering and Generative Models
Read
Uncertainty-Aware Language Modeling for Selective Question Answering
Presenting uncertainty-aware LLMs capable of estimating uncertainty with every prediction.
Read
Probability vs Confidence
Why a model’s output probability is not always a reliable guide to whether its answer is correct.
Read
Robust and Trustworthy Deep Learning (part 3): Themis AI
Themis AI's cutting-edge technological advancements in robust and trustworthy deep learning
Read
Robust and Trustworthy Deep Learning (Part 2): Uncertainty
Themis AI's cutting-edge technological advancements in robust and trustworthy deep learning
Read
Robust and Trustworthy Deep Learning (Part 1): Bias
Themis AI's cutting-edge technological advancements in robust and trustworthy deep learning
Read
Preliminary Steps Towards Risk-Aware Image Generation
Using Capsa to automatically evaluate the quality of images generated by Stable Diffusion
ReadNext step · Talk to the team
Bring one repeated decision. We’ll show you where Themis fits.