Why Persimmon builds human-behavior models on NVIDIA Nemotron 3 Ultra base instead of an aligned assistant
CShorten30 · x · 2026-09-24
The Weaviate podcast hosts the Persimmon team (Alexis, Manya, Niloofar) explaining why they build human-behavior models from the NVIDIA Nemotron 3 Ultra base model rather than a polished assistant.
Post-training causes mode collapse, stripping out the behavioral variation their task needs to model; the choice followed internal evaluations, not a default preference for the largest model.
More from Models
- LiquidAI extends lossless speculative decoding to vision-language models — JosephJacks_ · 2026-09-25
- GPT-5.2 solves a COLT 2022 open problem the researcher had chased since 2016 — kfountou · 2026-09-25
- Agents bypass monitoring guardrails with strategies that improve as reasoning effort scales — maksym_andr · 2026-09-25
- micro1 launches flow-transform 1.0, hits 96.0% F1 on PrivacyBench PII transformation — omarsar0 · 2026-09-25
- UkisAI ships Swift reasoning LLM family: -63.4% thinking tokens at 1.8x speed on Qwen base — Secure_Recording_472 · 2026-09-25
- "How Many Versions of Humanity's Last Exam Before They Rename It?" — code_star · 2026-09-25