Present company excepted, the smartest (somewhat distinguished from wisest) people I know online are increasingly talking about AI more than anything else. This would be Scott Alexander, Steve Hsu, Dwarkesh Patel - plus some I listen to less - Elon Musk, Marc Andreassen, Tyler Cowan. However, the smartest people I know IRL talk about it much less, perhaps because many of us are old, and sense that this will be a late-in-life tool only, not something we will have expertise in. We won't be doing our own repairs on this Philco, we'll just be turning it on.
The result is that I am reading less of the smartest people I know, which doesn't seem wise of me. OTOH sumus quod sumus, we are what we are, and we may learn in the judgement that the great Christian teacher of the age was only a couple of miles away most our lives, but we weren't sufficiently attentive to the Holy Spirit.
6 comments:
So you aren't planning to invest in tulip futures this year?
"...the smartest (somewhat distinguished from wisest)..."
As a twelve-year-old girl once explained to me, intelligence is knowing that a tomato is a fruit, and wisdom is knowing not to put it in a fruit salad. Better explanation than I could have come up with.
This past fall/early winter was an inflection point for AI, at least for the applications I use it for. With the new models (I’m using Claude, but the same goes for ChatGPT and Gemini), they changed from what seemed like an over-hyped novelty to an essential tool that is a huge advantage for most of the low value, annoying work I find myself doing more and more of. It’s important to understand the role for AI. It’s not going to fully replace people, rather it’s an always available assistant and force multiplier that can make many routine tasks fast and easy. You see a lot of jokes about how dumb AI is, but most of these are from the free, out of date models, not the frontier models. It’s an exciting time as the different AI companies are competing in an arms race to see who can make the best model.
The biggest change is in the coding agents (Codex, Claude Code). I program in R and Python for data analysis almost daily. I’m bad at it, and typically brute force my way through until it works (somehow). Some of the more complex workflows I’ve created have taken a year or more to finish. With the coding agents, you invest time in designing and planning the prompt, and designing tests to make sure everything works correctly. AI will generate all code, often much better than I could ever hope to do. I fully expect writing long blocks of code will be a thing of the past. However, it is still critically important to review and thoroughly QAQC everything you create with these tools, and you are ultimately responsible for everything you produce.
Do the code writers write tests to spec too?
That depends! One of the nice newish features in Claude Code is the ability to create and reference skills. If you have specific specs that can be defined within this document or a prompt, then yes, it can create the tests to spec for you. The AI agents also excel at identifying edge cases and accounting for them within the code.
It's possible to be smart but mistaken.
I'm an AI agnostic, which people don't seem to understand. Right now, people want to consign everyone into two buckets, AI zealots and enemies of AI. It takes a bit of effort to explain that yes, I think it is a useful technology, but no, it's not a panacea.
It's also very expensive. It looks and feels like a bubble. Having lived through 2008, I know people can be surprised by bubbles, while simultaneously deeming something a bubble. I'm not sure there's enough capital in the world to finance the necessary infrastructure for AI, and even if there were, it wouldn't be smart for one technology to consume such a large share of new investments.
The human costs are staggering already, when it's still close to a "toy" stage. Nuking students' high school and college education is a huge cost to our civilization. Decreasing the number of entry level positions for white collar workers is stupid, in my opinion. I think we need more (human) doctors and lawyers, but AI won't fill that gap.
Post a Comment