Liang Wenfeng, who started his career in quantification and is now all in AGI, has a different logic in his speech compared to those who speak incoherently about ecology. He said that AGI can be compared to climbing stairs. Last year's staircase was CoT, this year's staircase is Agent, and the next staircase is continuous learning. Continuous learning leads to the gradual singularity, and then to embodied intelligence. The embodiment is the last stop, and his reason is that for normal people, the demand is not for a computer, but for manpower. This sentence has fully outlined DeepSeek's technological priorities for the next three to five years. At the Agent level, he also has trade-offs. At present, the most important ones are Coding Agent and Universal Agent, with vertical scenarios such as finance and healthcare being given lower priority. The gap of the next generation model is continuous learning He said that AI doesn't lack taste and intuition now, what it lacks is the ability to continuously learn. Human beings can continue to learn, while AI has not yet. The next generation model must possess this capability, which is the fastest way to achieve AGI. But there are many things that make up this subject of learning, and the world has not yet found a good way. The commercial strategy is also the reason why DeepSeek insists on open source and ultimate cost-effectiveness, only making reasonable profits. We highly consider cost efficiency, so the price can be kept to the lowest possible level. Selling APIs, no sales, no customer service, users will come themselves. He never believes that big model companies can take away most of the profits in the AI industry. Open source is a part of restraint. He said that a software company's market costs billions of dollars a year, but once it's open source, it's gone. But AI is big enough to ultimately account for 10% of human society's GDP. To monopolize this benefit, one must be abandoned by history. The cost, time, user experience, and competitive gap are reflected in these three aspects. Cost first, a few months ago it was different. User experience is sticky, but not essential. Industry pattern, resource dispersion will inevitably converge The gap in AI in China mainly lies in resources. They believe in scaling, the larger the scale, the better the effect, but that's all the resources. There is no talent gap, they are all the same group of people. There are too many domestic model companies with scattered resources, and they will eventually converge. Each household only earns reasonable profits and doesn't need so many people to make a big model. Perhaps two large companies or two small companies are enough. Anthropic surpassing OpenAI is only a temporary phenomenon. In the long run, OpenAI and Google are likely to rise alternately. My opinion is that resource mismatch is the biggest waste of AI in China. If the same group of people are dismantled into dozens of companies and internal conflicts arise, convergence is bound to occur. Organizations and people, restraint is strategy. I have no intention of becoming the next super app, I don't like this thing. Don't lose watermelons for sesame seeds. Last year, everyone competed for Chatbot to grab C-end traffic, and this year they compete for B-end revenue. What they really care about is the AGI roadmap. A restrained organization has a restrained strategy. A company that is caught between KPI and GMV cannot say things like open source and only making reasonable profits. If your vision is to take more, you will lose first.