/TECHLei Li Group Presents InteractScience Benchmark and Four Papers at ICML 2026LL#1101Source Posts on XCluster sourcesLei Li@lileicsTECH#ICML2026 InteractScience builds a benchmark of 150 science demonstration-code generation tasks. Comprehensive unit tests for both code functionality of the generated code’s HTML DOM tree, and visual-semantic verification, using checklists that examine aspects e.g. num accuracy https://x.com/lileics/status/2075130850371363310/photo/1 https://twitter.com/lileics/status/2075130146558808131Combined views1.1K2 posts, first seen 7h ago11 likes2 saves
Lei Li@lileicsTECH#ICML2026 InteractScience builds a benchmark of 150 science demonstration-code generation tasks. Comprehensive unit tests for both code functionality of the generated code’s HTML DOM tree, and visual-semantic verification, using checklists that examine aspects e.g. num accuracy https://x.com/lileics/status/2075130850371363310/photo/1 https://twitter.com/lileics/status/2075130146558808131
Lei Li@lileicsTECH#ICML2026 [ICML 2026] Today’s fourth presentation — yes, busy day🫠 — 5 - 6:45pm Hall A, #904. We will introduce InteractScience, a benchmark for evaluating code generation for scientific demonstrations. Paper: https://arxiv.org/abs/2510.09724 https://x.com/lileics/status/2075130146558808131/photo/1 https://twitter.com/lileics/status/2074131115267215690