We are clearly at the beginning of a new tech epoch with AI. In many ways, the moment feels like the early internet - we can see a vast landscape of possibilities opening up, entire new categories of problems that can be solved, but in the near term, there are all kinds of implementation challenges.
While it’s important (and necessary) to solve those problems, it’s even more important not to be stuck to them. It’s very easy to build something “small” that feels satisfying, just because it’s new and possible at all. We are seeing a lot of that right now, “dancing bears” and party tricks that aren’t very scalable or useful.
The trick in this moment is to be ambitious to the point that your friends think you’re kind of nuts. It might be hard to imagine if you grew up with them, but there was a point of time where “the infinite bookstore” (Amazon) or “all of the world’s knowledge” sounded like a crazy, impossible dream. There used to be debates about how expensive search was, that it was an impractical idea, and lots of skepticism about eCommerce ever being widely adopted.
Right now, advanced AI models are hard to get access to, expensive, and slow to call. That will all change. In fact, once an idea has been discovered, which is hard, optimizing it for speed and cost is something that is both easier to do (so more people can help with it) and parallelizable, so it can happen much more quickly. We should expect that here.
The right thing to do is to pretend the performance and cost are already where they need to be. This is one of the “what if?” questions. What would you build once the LLMs are a few more generations better, and only take 10ms to call? What if you had something two or three generations smarter, that you could call not just once per user interaction, but thousands of times (which is how most large services work today - a web search or product page load is thousands of internal RPCs in parallel).
It’s hard to do this - it’s quite a leap of faith, and we often don’t get it right. But it’s essential when we are on an accelerating curve like this - you can’t just aim for what’s next, you have to be unreasonable and aim for what could become.
Hey I was just watching your interview with Kevin Scott at 'Behind the Tech' (I'm doing some research on Google Docs) and I wanted to let you know that your comment about life and work resonated with me. That idea for most people that 'work is equivalent to suffering'. The 'find the thing you feel kind of guilty about getting paid for, do the hell out of it' is a great mantra. Well said!
I found that that soon enough (writing about tech, I've got a BS in Computer Science) and I remember my father —he died sometime ago— always looked at me like saying: 'But you're enyoing your work, something IS wrong' :) I guess he always thought I would end my days at Deloitte or some place like that suffering each day of my life.
Thank God I didn't end like that. I'm sure in the end he agreed with you and me and is smiling from somewhere.
Good to tell this little story. Take care :)