TEM: Transformer Energy-based Models scale beyond diffusion without guidance
du_yilun · x · 2026-10-30
Researchers introduce TEM (Transformer Energy-based Models), claiming the first scalable contrastive energy-based model. Key claims: TEM outscales diffusion models without needing guidance, and shows better generalization, test-time scaling, and OOD detection than diffusion. A detailed thread accompanies the announcement.
Related event: TEM: Energy-Based Models Scale Up for the First Time, Beating Diffusion(2 posts)→
More from Research
- Randomly Disabling Qwen's Neurons: Coherence Collapses by 40% — flowersslop · 2026-10-30
- TWAS gets gene effect direction wrong 23-33% of the time, Nature Genetics study finds — anshulkundaje · 2026-10-30
- Hugging Face ran AI agents on 2,200 papers; some claims didn't hold up — evijit · 2026-10-30
- US spends $190B a year on R&D but results stay trapped in PDFs; Project Protostar wants to fix that — evijit · 2026-10-30
- AI-for-science benchmarking markets: single evals may cost months and millions, a durable niche — kenbwork · 2026-10-30
- pstack's /statechart skill ships with a mini TLC compiler that helped find real distributed-systems bugs — DavidKPiano · 2026-10-30