The Download: OpenAI’s turning point for math and a battery record
| Source: MIT Technology Review AI
Tags: OpenAI, mathematics, AI-agents, energy-storage, Danijar-Hafner, world-models
MIT Technology Review's daily digest covers OpenAI agents solving a major open mathematics problem while facing credit-attribution accusations, US battery installations hitting a record 20.2 GWh in Q2 2026, and Danijar Hafner's new venture migrating world-model agents into physical robots.
Details
The most consequential item in this digest is the OpenAI mathematics controversy: OpenAI claims its agents solved one of the most important open problems in mathematics, but the announcement is shadowed by accusations that OpenAI failed to credit researchers whose AI-assisted work influenced its solution. MIT Technology Review frames this as potentially a turning point — AI now seems necessary for progress on the hardest mathematical problems, but the resources required to run those models exist only at a handful of frontier labs, raising questions about where human mathematicians fit in. The second story covers US battery storage: Q2 2026 saw 20.2 GWh of new battery installations — enough for 600,000 homes' daily needs — driven by cheaper hardware and grid pressure from renewable additions. The piece notes different dynamics for grid-scale vs residential segments. The third piece profiles Danijar Hafner, who built world-model agents that master video games (DreamerV3) and is now migrating that approach to physical robots. He describes his new San Francisco company as continuing this long-horizon work on environments the agent hasn't seen in training. As a newsletter digest, this article is a pointer to three separate stories rather than primary reporting on any one of them.