This week

Potts calls a preprint's cross-model evidence weak at best

Stanford's Christopher Potts publicly questions a preprint's key claim, writing that the evidence for "subliminal learning" carrying across different models looks "weak at best." Potts is a natural language processing researcher at Stanford. Potts's October 26 post on X responds to the October 22 preprint "Tracing Subliminal Learning Through Pretraining." Subliminal learning is when one model is trained on data produced by another model and picks up the other model's traits through data that looks unrelated. The abstract states: "we find that transfer persists in both settings across differences in weight initialization, pretraining data order, pretraining corpus, model size, and architecture." Potts asked: "Are Figures 4B, A19B, and A26 really pictures of cross-model transfer?" Potts added: "The effects seem weak at best to me." Potts also noted that same-model transfer mostly appears early in training and cross-model transfer mostly late, which suggests two different mechanisms. These are Potts's methodological questions about the figures; Potts didn't rerun the experiments. Until the authors respond, cross-model transfer isn't a settled result (Christopher Potts on X, 2026-10-26; arXiv 2610.30884, 2026-10-22). How we checked →

Concept illustration on a peach background: at top center a black laptop whose screen shows a speech bubble and a large red dot, with two chains of arrows and paper sheets running down from it to the left and right; at lower left a black laptop with the same speech bubble and large red dot stands beside a tent card lettered "SAME MODEL", and at lower right a lighter laptop showing a speech bubble and a small red dot stands beside a tent card lettered "OTHER MODEL", with a hand holding a magnifying glass over that lighter laptop and a large pair of black quotation marks in the upper right. Across the bottom lies an open printed document lettered "ARXIV" at the top of its left page, filled with gray text lines and blocks.

Source and checks

This item ran in the brief of Oct 28, 2026, alongside 15 other items. How we checked it, the quote checks and what was still missing that day are on that issue's How we checked page.

Cite this item

SecondSource (2026). "Potts calls a preprint's cross-model evidence weak at best." https://secondsource.io/en/expert/2026-10-28-1/. Accessed access date.