Fetching from the wire…
01
02
03
04
05
06
07
08
Agents2026-08-18 · source-backed
A ten-author team including Emilien Dupont, Josh Alman, and Virginia Vassilevska Williams reformulated the combination-loss optimization behind the laser method, applied modern ML-based optimization, then used AlphaEvolve as a final refinement step (arXiv 2608.16884). Previous best: 2.371339. The division of labor is the story: humans reformulated the problem, the evolutionary coding agent squeezed out the last digits. One of the cleanest documented cases of an agent contributing a checkable delta to a headline theoretical-CS result rather than a benchmark score.
Each link below shares sources, entities, or timing with this story.
Google DeepMind released AlphaEvolve / Shared entity: AlphaEvolve / Shared topic / Earlier coverage
Linked by a graph relationship (Google DeepMind released AlphaEvolve); both cover AlphaEvolve; overlapping topics (agent, alphaevolve, coding).
Google DeepMind released AlphaEvolve / Same source domain / Shared topic / Tension
Linked by a graph relationship (Google DeepMind released AlphaEvolve); reported by the same outlet (arxiv.org); overlapping topics (agent, delta).
Google DeepMind released AlphaEvolve / Same source domain
Linked by a graph relationship (Google DeepMind released AlphaEvolve); reported by the same outlet (arxiv.org).
AlphaEvolve uses Gemini / Shared topic / Tension
Linked by a graph relationship (AlphaEvolve uses Gemini); overlapping topics (agent, best, coding); pushes against this story (but).
Google DeepMind released AlphaEvolve
Linked by a graph relationship (Google DeepMind released AlphaEvolve).
Linked by a graph relationship (Google DeepMind released AlphaEvolve).
Linked by a graph relationship (Google DeepMind released AlphaEvolve).
AlphaEvolve uses Gemini / Same source domain / Shared topic
Linked by a graph relationship (AlphaEvolve uses Gemini); reported by the same outlet (arxiv.org); overlapping topics (agent, benchmark).