Models are getting better at long-horizon tasks. So why don't they help much when you're planning clinical development or thinking through a launch?

Reinforcement learning may boost LLMs today, but it cannot deliver safe, long-term intelligence. We argue for System 2 learning instead.

Elicit develops LLM-based auto-evals to balance scale, trust, and flexibility, ensuring reliable scientific reasoning at superhuman speed.

We evaluate an automated approach for catching hallucinations in paper abstracts, aiming for consistently more trustworthy results.