Research

Bridging the gap between hypothesis and production.

World Economic Forum

How to make AI work in your enterprise through integration, not silos

Enterprise AI can be tricky but forward-deployed engineers can sit alongside teams for better results.

GenEval 2: Addressing benchmark drift in text-to-image evaluation

A new benchmark, GenEval 2, improves coverage of primitive visual concepts and higher degrees of compositionality, which we show is more challenging for current models.

Invisible Technologies

How much of your data can suck? Thresholds for domain performance and emergent misalignment in LLMs

This paper investigates the impact of incorrect data on the performance and safety of large language models (LLMs), specifically gpt-4o, during supervised fine-tuning (SFT).

Use of simulation to train AI for swarm based underwater behavior – Lessons learned from Talisman Sabre 2025

Exploring Resilient and Autonomous Artificial Intelligence Technology (RAAIT) trials to collaborate data feeds in the experimentation process.