@MaxForAI: Just now, Zhipu officially announced a $5 billion financing round to rival Silicon Valley AI Labs, going all-in on RSI!…

Just now, Zhipu officially announced a $5 billion financing round to rival Silicon Valley AI Labs, going all-in on RSI!! On the evening of September 13, Zhipu announced the completion of approximately $5 billion in equity and debt financing, including about $2 billion in share placements and about $3 billion in convertible bond issuances. More noteworthy than the financing scale is where this money will be spent next. According to the announcement, about 60% of the net proceeds will be used for the next-generation GLM foundational model, a fully self-training system, as well as large-scale training, production inference, computing resources, and related technical infrastructure. Another 15% will go toward business expansion, strategic investments, and potential acquisitions, with 25% allocated to working capital and the like. The most noteworthy among these is Zhipu's first explicit disclosure of "Fully Self Training." By Zhipu's definition, the next-generation GLM will be trained in the environment built by the previous-generation GLM, ultimately forming Recursive Self-Improvement, or RSI—a recursive self-improvement closed loop covering three directions: data self-generation, environment self-creation, and infrastructure self-optimization. Specifically, the model will participate in generating and screening its own training data; Agents will collect real tasks, generate validators, and continuously create new training environments. Even operators, kernels, scheduling, caching, and the Serving Stack are starting to involve the model in their optimization. This timing is quite intriguing. Just yesterday, the overseas AI community suddenly saw a surge in the spread of news claiming "Google DeepMind has achieved RSI." Although there is currently no sufficient public evidence to prove that Google has truly achieved complete RSI, this concept has recently clearly shifted from science fiction discussions into the real agenda of cutting-edge labs. Previously, OpenAI had also publicly discussed similar directions multiple times. OpenAI Chief Scientist Jakub Pachocki mentioned in a recent article that AI is increasingly participating in AI research itself, and in the future, it may enter a stage of recursive self-improvement. So, what truly merits attention in Zhipu's financing this time may not just be the addition of another $5 billion in ammunition. Rather, it's that OpenAI is publicly discussing RSI, rumors are wildly circulating overseas that Google has touched on RSI, and now Zhipu is directly incorporating Recursive Self-Improvement into the R&D roadmap for its next-generation model, preparing to invest billions of dollars to build the model, training environment, and computing infrastructure for it. In the past, everyone was competing on who could train the strongest next-generation model. Now the competition has begun on: Who can be the first to have this generation's model participate in creating the next-generation model. The paradigm has shifted, folks.