News · 11 June 2026
AI pre-training hides new physics discoveries
Photo by Shubham Dhage on Unsplash
This new study shows AI can struggle to recognise signs of new physics in the universe if it’s trained on old models. Researchers from Princeton University and the Flatiron Institute found this ‘negative transfer’ when AI was pretrained on simulations of the standard cosmic model. It means AI might interpret clues for massive neutrinos as existing features, missing genuinely new physical laws.
Reported by Open Access Government · How we write briefs