Weaviate Podcast #145: Humans& Researchers on Persimmon, a User Model That Predicts Human Turn-Taking
CShorten30 · x · 2026-09-21
The 145th Weaviate Podcast features Alexis Ross, Manya Bansal, and Niloofar Mireshghallah from Humans& on making AI understand humans better.
The centerpiece is Persimmon, a user model designed to predict how people respond in multi-turn conversations, including natural turn-taking and multi-person interaction. Notably, the goal isn't a more human-sounding assistant — it's matching the distribution of human behavior, including pushback, confusion, and frustration.
The guests argue that while AI models are advancing rapidly in coding, math, and classification, they still underperform at more human tasks like teaching and writing. Worth a listen for anyone interested in human-AI interaction and user modeling.
Related event: Weaviate Podcast #145 Discusses Persimmon User Model(2 posts)→
More from Models
- PrismML ships Bonsai 2 27B: 9x smaller ternary model keeps 98.2% of full-precision performance — thione · 2026-09-21
- Recap claims Google shipped Gemini 3.8 Live and DeepSeek released V4.1-Flash MoE — thione · 2026-09-21
- DeepSeek launches V4.1-Flash: 552B MoE with 8B active params, 1/4 the KV cache — thione · 2026-09-21
- Recap: OpenAI introduces disclosure framework, publishes six concerning-behavior reports — thione · 2026-09-21
- Kev: tiny Jev-like family of decision models built on Qwen3.5 hits Hacker News — petrusenko_max · 2026-09-21
- ChatGPT refuses to find a pirated ebook; DeepSeek just hands over the link — flavioAd · 2026-09-21