Jev fills the general-purpose classifier gap: labels in code, no extra LLM call
alexcovo_eth · x · 2026-09-19
A discussion about Jev, pitched as the missing general-purpose classifier. Responding to "classifiers always existed," the author concedes BERT, DeBERTa, GLiNER and fine-tuned heads were all specialized — changing labels meant another training run, and faking it with LLM JSON output was slow, expensive, and poorly calibrated.
Jev's pitch: labels defined in code, typed answers with confidence scores, no extra LLM call. The argument isn't that classification is new — it's that a classifier this intelligent, fast, and cheap is new.
More from Models
- PAW edges out Gemma e4b in early tests, with reliability as the real win — yuntiandeng · 2026-09-19
- GPT-6 Astra cracks a WWI German radio cipher, HN discussion — petrusenko_max · 2026-09-19
- DeepSeek 4.1 Flash surprises user by generating a QR code for phone auth unprompted — mariofilhoml · 2026-09-19
- Leak: OpenAI Pro 20x Signups Closed to New Users for 9 Days on Compute Crunch — haider1 · 2026-09-19
- User Spots Unconfirmed OpenAI Model With Blue Team / Red Team Setup — lxfater · 2026-09-19
- Dev Reviews 287 Open-Source Jev Projects, Picks 20 That Show What It's Good At — chenrongwei · 2026-09-19