Research updates, company news, and perspectives from the Eternis team.
We trained an 8B-parameter model that surpasses all published baselines on open-ended forecasting, including models 10-15x larger. Here's how context building, calibrated reward design, and modified GRPO got us there.
Our thesis on why the next generation of AI must be able to continuously refine its own architecture, and what that means for forecasting, sovereignty, and the future of intelligence.
Maximizing GRPO post-training under annotation budgets. Focusing on the hardest examples yields up to 47% performance gains.
