Researchers question whether LeJEPA's method scales beyond ImageNet-sized data
gabriberton · x · 2026-11-07
In a technical exchange, gabriberton argues there's no evidence LeJEPA scales well to large datasets: its public weights are trained only on ImageNet-1k/21k, versus DINO v2's 140M curated images and DINO v3/siglip's billions. He suggests the convincing test would be proper scaling-law plots showing how each method improves as data grows — pointing to a common gap between small-benchmark results and large-scale performance.
Related event: Researchers Question LeJEPA's Scalability to Larger Datasets(2 posts)→
More from Research
- Science's Last Exam: Best of 8 frontier models scores 37.5/100, all zero on end-to-end research — geoffwolfe · 2026-11-07
- New Benchmarks DecoyCount and ScienceVision Probe Whether Vision Models Fail at Seeing or Reasoning — AndrewDai · 2026-11-07
- BlancheMinerva argues withholding pre-camera-ready collaboration harms scientific practice — BlancheMinerva · 2026-11-07
- Aaronson-Ambainis conjecture caps query complexity, not super-polynomial quantum speedups in computation — thomasahle · 2026-11-07
- Flower Computers publicly releases encoders, billed as the fastest way to embed text semantically — matdryhurst · 2026-11-07
- HKUST and Tencent unveil Möbius: unified motion-language model beats 3B specialists at 0.6B — bdsqlsz · 2026-11-07