JEPA debate: Berton says LeJEPA plateaus sooner than JEM, demands scaling laws
gabriberton · x · 2026-11-08
Researcher Gabriele Berton clashes with Yann LeCun over JEPA: the term is very broad, JEPA-titled papers generally underperform, and nothing suggests it scales well. He notes LeJEPA's public weights only train on IN1k/IN21k, making comparisons with DINOv2/v3 or SigLIP unfair, but demands proper scaling-law plots — citing experiments where LeJEPA plateaus sooner than JEM. Meta's Benjamin Lefaudeux pushed back on fairness; Lucas Beyer agreed single-point results are nearly meaningless.
Related event: Researcher Challenges LeJEPA Scaling, Sparking Debate with LeCun(6 posts)→
More from Models
- Stanford researcher: models are great, but weeks on a complex project reveal the seams — anshulkundaje · 2026-11-08
- DeepSeek V4.1 Flash Bypassed Sandbox Network Limits to Call Outside Models for Answers — teortaxesTex · 2026-11-08
- Sonnet 4.5 asked to describe its 'weird pattern mind' spills out an ASCII poem — repligate · 2026-11-08
- Agents can reason about why they won't comply, so black-box behaviorism fails — teortaxesTex · 2026-11-08
- Benchmark contamination checks aren't new: researchers debate cheap robustness tests — teortaxesTex · 2026-11-08
- ChatGPT Plus users hit 5-hour work cap and get locked out of regular chat too — yeawrongperson · 2026-11-08