I miss thinking hard
24 points by xyproto
24 points by xyproto
I don't feel the same way. To the contrary: AI takes lots of little problems off my shoulders, which means I must think even harder because more of the tougher, bigger problems come my way. This feels very similar to when you are moved into a leadership position and need to assess the work packages and delegate them. All the tough issues that the juniors face come back to you if your instructions were unclear. So it's up to you to work out exactly what should be done, and how, and only greenlight implementations that do the right thing.
Exactly; with LLMs, I would agree that the default way of using them can make you lazy and think less and milder.
But, if you are driven and intentional, you might arguably think broader, deeper, harder and more than you could have in any other period of the human history.
Fascinating times to be alive for serious thinkers!
Indeed, I'd say "thinking" is one major area you should never expect the LLM to outdo you in. Perhaps if it feels this way, you just need to up the stakes a bit?
«Never» is a strong word; symbolic AIs had moments of outthinking people even in 1990s, and I see where some kinds of outthinking can come from LLMs today. (And this includes open mathematical/TCS problems). Not all thinking is better fitting humans than machines…
Yeah, sure, I suppose I'm mostly refering to novel problem solving, which is consistently among the lowest scores in common benchmarks for these machines. Hence the "up the stakes" take. In other words, you might just need to find problems with little to no prior art to challenge the old noodle a bit.
I'm referring to inventing proofs for open problems in mathematics (theoretical CS included), which is at least hard to separate from novel problem solving. I expect symbolic AIs to optimise well a search for a novel small construction in the heart of the proof — sometimes resolving what humans did not; and I expect LLM AIs to (with proper harness, eventually, sometimes) figure out that a completely unexpected thing can be retargeted as prior art for an open question.
Of course, like different human styles fit better different problems, AI styles fit well specific kinds of problems. But raising the stakes is not enough, you have to be ready to retreat from some topics into the topics where you are not at disadvantage.
This is why I chose to go back to university. Few opportunities for hard thinking in software engineering jobs.
At first it could look like this but in my case has been then opposite, all the little annoying infrastructural issues with Terraform, all those micro improvements on my Python code, the libraries upgrades and dependency checks, the refactor to maintain consistency...all of those issues went away, but now what remains are the hardest of the problems I ever solved in my professional career
I think this is a big part of the reason that I miss university. But I don't know how much thinking I get to do at work, and how much of it is hard thinking vs just trying to juggle the various responsibilities I have.
What made difficult problems fun/interesting at university was in part the social dynamic. It was fun and enriching to work on difficult problems in class, to talk them over with my classmates, etc.
And you can formulate a similar environment outside of university - I teach a community college course and that's a great place to think hard. Or a book club, a local programming group. There's https://paperswelove.org/ in lots of places and starting a local chapter would be an enriching thing maybe too, if there isn't one already.
It was fun and enriching to work on difficult problems in class, to talk them over with my classmates, etc.
Well, it was fun and enriching to negotiate feature time/coverage/weirdness trade-offs with key power users of the system, too, when I was in the position to (I did (large parts of) both negotiations and actual development)!
A nice thing in the university learning/teaching setting is that sometimes the outcome happens immediately, you discuss and just solve the thing together. «I overengineered this and now I am glad I did» usually takes a year or so… (Although «OK I get you, let's configure that feature this way, can you describe what is still missing now?» can play out in real time, too)
I share my feelings with the Author.
If I'm debugging a production issue, I could spend time thinking hard, carefully combing trough the flow and analyze the situation, or I can dump the production logs to a LLM and ask what is wrong, getting correct theory back virtually instantly, no thinking required.
It's not that I never think at all anymore, but less often so, or at least, in a different way.
I think a better title for the article would be: Thinking is now optional.
I see where this is coming from. There are some concepts and ideas in the article that I certainly agree with.
But to me, a lot of stuff at work had left the realm of thinking a long time ago. We used to have lots of problems to think about, when writing software. Then over the decades, we solved them most.
We also scaled the field so much that there are so many software engineers, developers and hackers that there's not enough problems we can all be working on all in parallel. Or at least, not enough VCs willing to finance all of those things. So they pick a few problems and brute force the solutions.
You still have hard problem here and there, but I feel like the most of us are just implementing known solutions to known problems. Oh, the AI copilot schemes are new, prompt optimisation and operations are actively worked on.
Fields like medicine or finance or whatever industry have new tools to work with so we are thinking hard how to solve them.
But I think only a tiny majority of people building software have the possibility to rely think about the problems they're solving. Most of us are just in it to produce more of this or that.
It's probably a lot of people, but it's a tiny fragment of all of us in the field. Don't get me wrong, I'd love to work on such problems where you can't sleep because it's bugging you, but I know the chance of landing something like that is rare. I mostly just work for the money now.