Some takeaways from a chat with (, a former colleague in Silicon Valley and the 20th employee at a certain AI Startup:

1. OpenAI's technical strength remains the benchmark level. There were widespread rumors about talent loss, but it hasn't affected the core.

2. The impact of GenAI on traditional industries has just begun. They are already working on related businesses. This is a long journey, and only 1% has been advanced so far. The speed won't be very fast because domain knowledge transfer and context building take time, and communication costs are significant.

3. Tech builders in Silicon Valley are generally very optimistic about the future prospects of AI because they are actively delivering new scenarios. They are also very optimistic about OAI, so he was shocked by the controversy over OAI's cash flow, not expecting Wall Street's suspicion of the AI Bubble and large capital expenditures to be so severe.

4. He doesn't have an "extremely bullish" view on Google. First, because AI is a tug-of-war with traditional businesses; second, because he thinks Gemini isn't that user-friendly.

5. Regarding Gemini, he dislikes the fact that context following is average, but that doesn't mean Google can't develop better context following. One speculation is that the company internally prepared hundreds of models, selecting the one with the highest benchmark score to release, aiming to suppress OAI's narrative and boost stock prices.

6. About the future development of technology, he still strongly believes in the views from Ilya's recent podcast—that the future will shift towards an era of research.

7. Pre-training is almost done; subsequent is post-training, and computing power will be continuously needed.

8. Claude Code remains his favorite vibe coding tool, and GPT is still the model he uses the most.

9. ), gossiping, doesn't like or have high regard for Alexandr Wang or the previous Scale AI company, so he thinks Meta's acquisition was another waste of money.