Meta-acquired AI chief's three-hour rant: Treat agents like factories, skip the hype tools, and win by deciding what not to build

Ji Yichao, Chief Scientist of Manus, which was later acquired by Meta, shared his experience in the AI field over the past decade in a three-hour interview. He discussed his journey from Tokenize to LSTM to Transformer-related applications and his experience in making AI browsers. Ji Yichao highlighted the success of Manus, which solved problems that Cloud Providers and Model Providers couldn't, by creating a good toolkit that allows Large Language Models to automatically plan things. He also mentioned the importance of optimizing AI Agents, similar to manufacturing, and the need to decide what not to do in product planning. The interview touched on technical concepts such as Model Context Protocol and code execution environments, which can improve the efficiency of using AI agents with external systems. Ji Yichao cited influential papers, including FLAN-T5 and Word2Vec, which have contributed to the progress of AI in natural language processing.