Keep pulling the thread on Demis Hassabis.
Reasoning systems based on foundational models are capable of achieving a gold-medal standard at the International Mathematical Olympiad.
Aletheia is a math research agent that iteratively generates, verifies, and revises solutions end-to-end in natural language.
The Aletheia system generated a research paper, designated Feng26, without human intervention to calculate structure constants in arithmetic geometry called eigenweights.
The Aletheia system was used in a human-AI collaboration to produce a research paper, designated LeeSeo26, that proved bounds on systems of interacting particles called independent sets.
The Aletheia system conducted a semi-autonomous evaluation of 700 open problems from Bloom's Erdos Conjectures database.
The Aletheia system autonomously solved four open questions from Bloom's Erdos Conjectures database.
The Aletheia math research agent is powered by an advanced version of Gemini Deep Think.
The Aletheia math research agent utilizes a novel inference-time scaling law that extends its capabilities beyond Olympiad-level problems.
The Aletheia math research agent uses intensive tool use to navigate the complexities of mathematical research.
The prompts and model outputs for the Aletheia project are available at https://github.com/google-deepmind/superhuman/tree/main/aletheia.