Holding two thoughts
I’m on a bit of summer vacation this week, so not as much to report on engineering. I have been watching and listening to all of the “AI is a bubble” talk, which I think is largely noise. This is a large, complex change - we are inventing a whole new form of computation and probably cognition! It’s going to be messy, and no single narrative will likely capture all of the truth - neither “it doesn’t work and is over invested” nor “it will instantly change everything”. It’s more like “this is really interesting, has a lot of promise, some folks are figuring out really productive things to do with it (read Ethan Mollick if you haven’t already), but we don’t fully understand how to best use it and we have lots and lots of technical work to do to optimize, secure, and develop it”. But that’s no fun to put in a headline.
I’ve been spending the week in Western WI. It’s a good antidote to being in Silicon Valley. It’s gorgeous here this time of year - a few pictures below, of my friend Dan, his farm, my barn, and some wonderful fresh produce. AI isn’t as much top of mind here - folks are curious but haven’t used it much, which I find interesting, since it’s a very easy thing to try (Ethan talks about needing to spend 10 hours with it to really understand it, and I agree - if you haven’t done that yet, you should!).
But it’s creeping in, in useful ways. Yesterday my wife and sister and I were going through old family recipes for a cookbook. Lots of handwritten stuff to transcribe. It’s just miraculous to be able to hand an image over and say “please transcribe” and get a perfect rendition back in 5 seconds or less. My sister was also working on a survey for a historical project and struggling with the questions to ask. A quick query got her 10 good ones that she could tweak to make her own.
These seem small, but they add up over time. I remember when no one thought ecommerce would work because it wasn’t “convenient enough”. I think I am probably getting at least 500 if not more packages annually from Amazon at this point (yeah, I know). I remember when search was kind of useful but not something you used all the time. This week I was contemplating that, if I lost my phone kayaking, it would be fairly hard to find the apple store to buy a new one, without the phone to guide me.
Things creep in. AI will too. And it’s always easier to optimize something once it’s working than to build it in the first place. I’m very confident the costs and value will align before long, and I’m very confident that this is a useful tool, even in its current form, that I will find many uses of.
Wisconsin pictures:
I had spent some time on a similar idea, not just scanning old written recipes but then using AI to convert it into a machine-readable format which could be fed into digital recipe programs. AI is good at transferring styles like taking text and laying it out into a JSON format.
Thanks Sam! I appreciated the balanced nature of this post. I've had similar experiences where AI "just worked" and was magical, followed by disappointing experiences on a task where I thought it would excel. Overall, really optimistic and I appreciate your perspective and comparing it to how development changed with the web.