millionways 发布 Thorsten-4，基于 7000 人四年纵向访谈数据训练
7000 人四年访谈数据训练出来的对话理解模型，验证方式是拿预测对照真实行为，不是刷基准分。做客服机器人、心理支持类智能体的团队可以直接看 demo，思路和主流爬语料训模型完全不同。
millionways 推出模型 Thorsten-4，其训练基础来自专业心理学家在四年内对超过 7000 人进行的重复访谈研究。团队将 Thorsten-4 的预测结果与这些人后续的实际行为进行对照验证，而非仅依赖网页上可抓取的公开数据。
该模型的能力定位是分析人际沟通：理解对方为何犹豫、预判其可能的回应方式。官方设想的使用场景是让智能体在处理困难对话时具备这类判断力，例如客服、心理支持类对话代理。Thorsten-4 可以与企业现有 AI 工具并行协同工作，不需要替换既有系统。
这类纵向访谈数据的构建成本远高于常见的网络语料训练路线，也难以被其他厂商复制。在普遍以基准分数衡量模型的行业里，用真实行为对照来评估对话理解能力是一种较少见的验证方式。
Expert psychologists interviewed 7,000+ people repeatedly over four years. The team then checked Thorsten-4's predictions against what those people actually did.
I'd like to see more AI evaluated this way. An agent handling a difficult conversation needs some grasp of why the person is hesitating and how they might respond.
@millionwaysai built Thorsten-4 around that research. The interviews aren't data you can scrape from the web. The model analyzes communication and can work alongside the AI tools companies already use.
Details / demo: https://t.co/9OotKvdqTE
This report is categorized under AITOP's "Models" desk, focusing on US frontier AI labs and Silicon Valley technology leaders. Within our cross-regional analytical framework, it directly maps to AI developer toolchains, coding agents, and engineering workflows and US-China foundation model benchmarks and commercialization comparison. Key entities and technical topics under continuous tracking include Thorsten-4, millionways, 对话理解, 心理访谈数据.
Source attribution: curated from kimmonismus (@kimmonismus).
