AIBuildAI-2.5
An autonomous modeling system based on large language model agents. It introduces three components—LLM-guided tree search, resource-aware scheduling, and cost-aware model routing—to enable autonomous end-to-end modeling from data processing through model training.
Timeline (last 90 days)
AIBuildAI-2.5 is an autonomous modeling system based on large language model (LLM) agents.
AIBuildAI-2.5 ranked first on MLE-Bench, achieving a medal rate of 73.3%.
AIBuildAI-2.5 outperformed a strong baseline system on six autonomous AI research tasks from AIRS-Bench.
The core innovation of AIBuildAI-2.5 is an LLM-guided tree search that introduces a judge role to score each candidate program along three dimensions: expected improvement, sufficiency of justification, and feasibility.
AIBuildAI-2.5 is equipped with a resource-aware scheduler that takes the current hardware resource status into account when launching training jobs.
An LLM judge can perform multi-dimensional evaluation based solely on the code text itself before program execution, forming a soft filtering layer that prioritizes the most valuable candidates for execution.
All Facts (5)
AIBuildAI-2.5 is an autonomous modeling system based on large language model (LLM) agents.
50%UnverifiedAIBuildAI-2.5 outperformed a strong baseline system on six autonomous AI research tasks from AIRS-Bench.
50%UnverifiedThe core innovation of AIBuildAI-2.5 is an LLM-guided tree search that introduces a judge role to score each candidate program along three dimensions: expected improvement, sufficiency of justification, and feasibility.
50%UnverifiedAIBuildAI-2.5 is equipped with a resource-aware scheduler that takes the current hardware resource status into account when launching training jobs.
50%UnverifiedAn LLM judge can perform multi-dimensional evaluation based solely on the code text itself before program execution, forming a soft filtering layer that prioritizes the most valuable candidates for execution.
50%