PhAI Labs trains research agents to improve hypotheses from their own work
Co-founder Ling Yang's DiscoveryRSI trains research agents using evidence and revision histories; its literature-match result does not show that researchers used the ideas.
By RuntimeWire Staff · Published
Primary source: PhAI Labs
Why it matters
DiscoveryRSI makes the research process itself part of model training, but its reported literature matches show correspondence, not scientific influence or validated discoveries.
PhAI Labs co-founder Ling Yang (@LingYang_PU) is betting that research agents can improve by learning how scientific hypotheses develop. The lab's DiscoveryRSI technical report describes a system that saves the evidence, decisions and contributions behind successive revisions, then uses those records to train a model and adjust how agents conduct later investigations.…
When you purchase through links in our articles, we may earn a small commission. This doesn’t affect our editorial independence.