ActICL trains multimodal AI agents to choose in-context examples
Its developers report an 11.6-point gain over similarity retrieval and a 4.4-point gain over their strongest trained baseline.
TLDR
An ActICL developer says the system trains multimodal AI agents to search a bank of demonstrations, inspect candidates and choose examples within token budgets. Across seven benchmarks and three backbones, they report a 4.4-point lead over their strongest trained baseline while using about 3.5 demonstrations on average. They also say contexts built by a 9B ActICL agent improved GPT-5.6-Sol by 3.5 points over standard retrieval.
Combined views
—
1 Source, first seen ago
