I've seen a lot of discussions about Xinke in the comments. My observation is that AI as a direction for productivity is absolutely fine, but also because the evolution speed of AI is too fast, apart from that 0.1% of truly top-tier AI scientists, the rest might become collateral damage... During my recent trip to the Bay Area, I felt this particularly deeply; the anxiety of big tech programmers is almost written on their faces. Microsoft has laid off over 15,000 people cumulatively by 2025, accounting for about 7% of its total workforce. The darkest humor is that they once established an 800-person "Efficiency and Cost Reduction Group" specifically to study how to use AI to save labor costs, and in the end, this group itself was cut by 200 people... Meta, on the other hand, is both poaching talent at exorbitant prices (Matt, born in 2001, was just poached this August with a $250 million package) while maintaining a culture of "rank and yank," with quarterly eliminations, and is expected to lay off about 5% by 2025. In the coffee shops along the streets of the Bay Area, you can almost hear layoff news everywhere. Another harsh reality is that PhDs from my cohort were all based on visual recognition machine learning back in the day. But this generation has completely switched to large language models and multimodal approaches. The paradigm shift has directly caused my classmates' packages to lag far behind the new PhDs who graduated in the last two years. The tide of the times cannot be stopped by anyone. Additionally, the first validated track for AI is actually vibe coding. Writing code is inherently structured input + verifiable output, and it can also be highly globalized, making it the easiest field to be replaced by AI. Cursor and Cognition are both valued at around $10 billion, becoming the most direct use cases of this wave of AI. Every generation feels like they are at the peak of the wave, but in reality, we are all just bubbles in the tide.