AIScientistTwo reportedly improves on human baselines in 86 of 107 machine-learning problems
A post describing a Google paper says ScientistTwo can improve a human method, use its own discovery as a new baseline and improve it again.
TLDR
According to a post describing a Google paper, ScientistTwo autonomously improved 86 of 107 human-defined machine-learning problems based on ICLR, ICML and NeurIPS papers. The post cites an 80.4% success rate and a 25.2% average relative improvement over the original human baselines. It describes a research loop that proposes ideas, tests them, removes what does not help and uses reviewer feedback to start another round.
Combined views
1 Source, first seen 3h ago
ScientistTwo reportedly improves on human baselines in 86 of 107 machine-learning problems
A post describing a Google paper says ScientistTwo can improve a human method, use its own discovery as a new baseline and improve it again.
TLDR
According to a post describing a Google paper, ScientistTwo autonomously improved 86 of 107 human-defined machine-learning problems based on ICLR, ICML and NeurIPS papers. The post cites an 80.4% success rate and a 25.2% average relative improvement over the original human baselines. It describes a research loop that proposes ideas, tests them, removes what does not help and uses reviewer feedback to start another round.