Post

Quitting programming as a career right now because of LLMs would be like quitting carpentry as a career thanks to the invention of the table saw.

Agreed. But to further the analogy it would be more like quitting carpentry as a career after the invention of a table saw that sometimes cuts straight lines and sometimes cuts crooked lines and sometimes just goes berserk and cuts your hand off. Even when given the exact same instructions. 🫠

I've gotten some useful summaries, snippets, and decent references from AI. It is like a slightly better Google in some cases (in other cases, worse). I worked w/a colleague a couple months back on a problem w/sorting derived, calc'd fields. We got 20 different answers in 20 prompt attempts. 🫠

None of the 20 answers provided or even led to a workable solution that covered all of our needs. We eventually had to request either a change of the specs or a change to use a SQL view vs. EF context. Our lead advised strongly against anything non-ORM. AI was less than useless in this case.

That kind of thing is why I like the power tools analogy: you need to learn a whole bunch of things to stay safe from those kinds of behaviors

It also feels like "here's a free power tool that cuts your arm off most of the time, but that's ok because there's a hugely expensive one that mostly works once you learn its quirks, and keep a surgeon nearby just in case"

What's hugely expensive though? Claude Code is $20/month and right up there with the best tools.

The $20/month plan tends to run out of tokens pretty quickly - I know a lot of people who have upgraded to $100/month or even $200/month plans purely to use Claude Code more aggressively

What frustrates me most of all is that people praising LLMs aren't transparent that this is *not* the experience you get from free ChatGPT or Gemini, or whatever. Maybe everyone is talking past each other, but IMO it's pretty basic that a tool not being free is worth being explicit about.

Yeah I have a strong theory that the (somehow huge) amount of people (on Bluesky specifically) who claim LLMs don't do anything only have experience of using the free ChatGPT 9 months ago or something.

I'll freely admit that my experience is solely with the free ChatGPT (and specifically the "stay logged out" version). I use exclusively free development tools, and I don't see why LLMs should be any different. And I don't want to deal with my interaction history becoming part of the context.

For most of these tools, paying for them is the thing that prevents them from training on your interactions

Wow, I didn't know that. So that's another reason I want transparency over costs. "Don't train on my interactions" is a non-negotiable baseline for anything non-trivial (*certainly* for code assistance) in my opinion.

Regrettably, the (lack of) credibility of AI companies makes it hard to believe such claims...

I wrote about that a while ago, I called it the AI trust crisis simonwillison.net/2023/Dec/14/...