Zhipu AI founder outlines Touch High plan for AGI research

Zhipu AI founder Tang Jie announced a 'Touch High' plan in an internal letter, prioritizing AGI research over short-term commercialization. The plan focuses on long-horizon tasks, autonomous agents, self-training, and extreme safety governance, with mechanistic interpretability as a key safety research area.

Zhipu AI founder Tang Jie told employees in an internal letter that the Chinese foundation model developer will launch a “Touch High” plan, focusing on AGI research rather than short-term commercialization, according to LatePost. Zhipu AI is the company behind the GLM series of large language models. Tang listed four technical priorities for the plan: long-horizon tasks, autonomous agent systems, fully self-training, and extreme safety governance. The letter also singled out mechanistic interpretability, which studies the logic behind model decisions, as a major safety research direction. LatePost https://www.latepost.com/news/dj detail?id=3637 , in Chinese