I coffee-chatted with several people after I moved to the Bay Area. Stanford PostDocs, Google full-time research scientists, Microsoft applied scientists, founders, investors…. And here are my reflections on my 1-month move here.
People are absolutely, desperately in need of the next AI natural force. The best engineers and elite researchers are looking for the ChatGPT 3.5 of 2022. But right now, it seems like there is no solution.
Coming to industry this time is different from last summer’s internship at Google. I felt more mature (in expectation), I felt more engaged in the product, I felt it more important to listen than to talk, I felt stronger with a Ph.D. degree, I felt more connected, and I realized that I am not really interested in money (yes, total compensation) but in the passion.
I am learning the difficulty of changing a non-AI-native product into an AI product. It is frustrating sometimes, feeling like finding a needle in a haystack: trying to find excuses to say why AI/agents are necessary in a given task. LLMs did make the one-person-company workstyle more common, but the traditional workflow still exists: big teams still have many meetings to work together, and still need to write reports and slides which I believe an LLM can make within 5 minutes and which definitely did not need to take 2 weeks of meetings and offline syncs to ensure delivery. So, with LLMs, how can people still work in a condition that used to have no AI at all? Just work for 5 minutes and use coding agents to write millions of lines of code and call it done? That’s one of the biggest questions we are all facing: What are we doing? How do we maintain and keep innovating on a non-AI-native product (there are millions and billions of products that are not AI-native at all)?
But it is indeed all the big tech companies’ problem now. It takes time and effort to pivot, and no matter whether it succeeds or not, it’s a huge investment: if it succeeds, great, but still a lot of money goes to the frontier-lab model companies to pay for their high-quality model tokens; if it fails, damn, we wasted money and time, paid for the best tokens, and lost customers. It becomes harder and harder to win, and it needs a lot of investment to ensure we have at least a foundation to win.
I am learning, more through a non-technical lens. I look at how people communicate, how a big corporation is operated, and how to XXX-maxxing in productivity, thinking, and product impact. It’s great that I have a chance to take so many steps back and look at the tech giant instead of researching alone. I hate to rely on other people and owe people things, but society doesn’t operate that way. People need to communicate, collaborate, and fight, and that is society.
I am always joking that Ph.D.s are a bunch of mean people: we only look at people through a technical lens and determine whether a person is “useful” or not. But society and life are not like that; there are more things than just being “useful”, and “useful” today doesn’t mean “useful” tomorrow. We need to listen to the trend, we need to pay more attention to the actual demands (instead of just reading social media posts and thinking that’s the trend), and we need to purposely stand in the right time, at the right spot, with the right audience, and make the right pitch. The actual demands come naturally.
Other Reflections on my Relocation to the Bay Area & Personal Updates
The Bay Area is still the central hub of funding innovation. That being said, the innovation is not necessarily happening here, but the money to motivate, fund, and nourish the innovation is here. You have to be physically here to feel like you are in the trend of AI. Boston is great, but it’s a good place to study and think, not a good place to talk with investors and meet founders, etc.
In the past month, I co-organized an open-source community, MatrAIx, with my long-term friend and collaborator Dr. Xiaomin Li. This community feels like a natural force. We just kept discussing it from very early this year, and one day, around mid-June, we both thought: how about… let’s initiate an open-source community to engage researchers who are interested in persona-driven agents for evaluation?
And then here it is: 160+ contributors (Ph.D.-level researchers and industry engineers), 3 groups in persona, environment, and application, 4 types of applications: survey, chatbot, website, and iOS/MacOS application, and 3 domains: e-Commerce, Gaming, and Healthcare. It’s all in one month. I am personally very, very proud of this achievement, and this is indeed a good lesson and plAIground for me to showcase what I learned from working in the industry.
Generally speaking, as my friends always ask: how’s the Bay?
My answer is always: Come and feel it :D
YH