Chatting with a good friend who did an AI PhD at Chinese University of Hong Kong and has worked on various AI models, key takeaways:

1. OpenClaw is garbage
2. A senior student from Beijing University of Posts and Telecommunications used Claude to write a project in 2 weeks and sold it for $30 million USD. Wants to start their own business but lacks ideas. The window of opportunity is this year.
3. Claude already has the ability to train itself. Other large models basically have self-training capabilities too, they're just not publicly disclosing it.
4. Video AI will have qualitative breakthroughs in the next two years. Editing AI is much harder than expected.
5. Particularly anxious about losing their job.