Encoder-style classification gets hot again: one multimodal BERT solved 1,000+ classification tasks

cwolferesearch · x · 2026-09-21

• cwolferesearch argues Jev is bringing structured/open-ended classification with encoder-style models back into focus, opening a new dimension in LLM workflows.
• He cites his own early research training a multimodal BERT to solve over 1,000 multi-class classification problems with a single model, showing the potential when trained and scaled properly.
• He notes Jev goes beyond basic encoder classification: scalable generalization to any classification space with efficient inference (likely an improved CLIP-style contrastive approach) and in-depth data-centric research.

Original post →

More from Models

• Open models hit 78.4% of Vercel AI Gateway tokens; DeepSeek and Moonshot spend top OpenAI — realmeetjames · 2026-09-21
• Reddit user: GPT 6 reasoning feels worse than 5.6 Terra, like GPT 3 flashbacks — FigInternational4873 · 2026-09-21
• Google launches Gemini 3.8 Flash and Flash Cyber: 54.9% HLE, 2.6x better vuln patches — AlienX__0 · 2026-09-21
• Open-source model Jev called out for not crediting Lava, its 8192-context prior work — nischay_twt · 2026-09-21
• Devin makes SWE-2 free across Cloud, CLI and Desktop until October 8 — HankYeomans · 2026-09-21
• Deployers told: benchmarks are a directional signal, not a substitute for your use case — evijit · 2026-09-21

Models channel →