Baidu Releases ERNIE Evolve Automated Research System, GPU Inference Throughput Improved by 33.7%

Baidu has released the ERNIE Evolve automated research system for improving AI training and inference: given a clear research goal, it selects research directions, modifies code, and runs experiments; only improvements that pass independent evaluation and review are retained, and the results of each round guide the next round. On public benchmarks, the system achieved SOTA on NanoChat AutoResearch and NanoGPT Speedrun; in ERNIE 5.5 Flash inference tests, deterministic GPU inference throughput improved by 33.7%, and single-request decoding speed improved by 14.9%. Baidu calls this a step toward recursive self-improvement (RSI), and plans next to extend the method to training systems, model capabilities, and evaluation.