Tang Jie, the founder of Zhipu, issued an internal letter: starting the Touch High plan and continuing to focus on AGI research rather than short-term commercial realization.

IT House reported on July 11 that according to a report by LastPost, Zhipu founder Tang Jie issued an internal letter today, explaining Zhipu’s understanding of AGI’s next competition. Tang Jie stated in the letter that Zhipu will continue to continue the so-called "counter-intuitive" route and launch the "Touch High (Touch High) Plan", that is, continue to focus on AGI research rather than short-term commercial realization. Tang Jie said that there are several mountains that must be climbed on the road to the end of AGI, and they are also where today's technology wave is the most turbulent: Long-range missions Autonomous agent systems Complete self-training Extreme security governance Among them, extreme security governance is particularly emphasized, and the Wisdom Spectrum Project invests tens of billions of resources to tackle mechanical interpretability, which means clarifying the neuron logic behind model decision-making and promoting the transformation of black box systems into transparent and explainable systems. IT House attaches the full text of Tang Jie's internal letter as follows: The big wave is coming - to every smart person and partner who cares about the future of artificial intelligence. Please allow me to use this article to talk about three things: who we are, how we view this era, and the strategic direction we have decided to devote our full attention to. (1) Who we are: “Essence, Counter-intuition, Focus” Zhipu has never been a company that chases trends. It grew out of a laboratory, carrying with it twenty years of methodology. This methodology can be summarized in three words: essence, counter-intuition, and focus - only if you think deeply enough can you dare to choose something counter-intuitive; if you choose something counter-intuitive enough, can you stick to it long enough. Looking back, almost every key choice we made seemed “counter-intuitive”. In 2006, we sat on the bench guarding the academic search system on a desktop computer because we figured out that behind it was "mining the evolution mechanism of the subject", which was worth ten years to answer; from 2021 to 2022, when "making machines think like humans" was regarded by most people as a crazy plan like a moon landing, we allocated resources, bet on hundreds of billions of parameters, and made GLM-130B - that is ChatGPT It was a full year and a half before the world exploded; and in 2026 1