OpenAI GPT-5.2 Pro extends amplitudes to gravitons
OpenAI says GPT-5.2 Pro helped extend a known single-minus gluon result into gravity, showing graviton amplitudes can be nonzero in a special half-collinear regime. The paper emphasizes that the model helped generate the conjecture, while the final result was still checked analytically by human authors.
This is a stronger signal than a flashy benchmark post: the model contributed to an actual research conjecture, then the work cleared conventional verification.
- –The technical win is narrow but real, which makes it more credible than broad "AI solved physics" hype.
- –The transfer from gluons to gravitons suggests these models may be useful for pattern-spotting across adjacent theory domains.
- –The paper is explicit that validation still matters; discovery may be getting faster, but proof remains human-auditable.
- –For AI labs, this is a compelling demo of model-assisted research workflows, not a replacement for domain experts.
DISCOVERED
71d ago
2026-03-18
PUBLISHED
71d ago
2026-03-18
RELEVANCE
AUTHOR
borowcy