I also experience this as a great loss, but honestly yes, I can't get excited about perfecting the craft of writing computer code now that there are AI systems that have done so. I'll always be sad about this! But the flip side is that I'm way more excited about what I can do with software, now that it's so much easier to write.
I experience quite the opposite. I'm quite excited about the value my skills are gaining in relative terms. An entire generation of potential competition is getting addicted to the subsidized corporate tooling. Voluntarily!
What I fear is the future career path for manual programming is fixing AI generated systems. I suspect it will pay very well because you will be finding and fixing things the AI driven work can't. It is going to be soul destroying as reading AI code is deeply unpleasant in the same way reading the articles is.
I never go by pay. I won't fix bad systems whatever their provenance because trying to understand poorly coded software is very unpleasant.
I go by interesting problems and I like being able to chose from more options due to lessened expertise competition. Why people voluntarily renounce expertise is beyond me, but they are and I welcome the increase in choices of engagement I have been experiencing.
I would never engage in uninteresting work at any rate of pay, and working with LLM output directly or mediated is quite uninteresting to me.
The code is ultimately the codification (ha!) of domain knowledge in precise, non-ambiguous ways. It has extrinsic, non essential complexity yes.
But including ever more complexity in the process, by depending on complexity-slinger systems that you don't fully control or even understand, is not the solution, in my view.
If you lose control over the precise, detailed specification of the domain knowledge that the code is, ultimately you are just a glorified business analyst. Besides, the inherent complexity of the implementation is not without essential complexity that needs mastering for security and performance reasons. Deferring that mastery to some tool is self defeating. The complexity doesn't disappear. You're just shuffling it around and when it leaves your purview you are merely creating opportunity for someone else to capture that value.
Ultimately yeah the vision of being an LLM jockey who doesn't know code might be viable or even numerically dominant in the market in future. Why would you descend the value chain though, and voluntarily, that I don't understand. Competing for volume isn't historically in the interest of the autonomous worker. Maybe I'd feel differently about it if I owned capital and had some corporate strategy behind it. As an independent actor, I don't see it favoring me.
AI is a very useful tool for finding problems in codebases.
I'm very far from an "AI is going to take everyone's job" doomer, but I really do think "there will be programming jobs that don't involve using LLM based tooling" seems pretty unlikely.
The future is open models running on your device. No subsidized corporate tooling required. There will never be a reckoning for the people who are letting their coding skills atrophy.
Yeah I know it's not obvious. And very lucrative and very freeing, having this sort of knowledge and leveraging it.
Especially after clients have suffered the new found arrogance of non-experts riding on LLM output.
Look, let me be perfectly clear: if you're getting what you want out of your job and your personal projects, hobbies, whatever you do with your time, awesome! If you feel like the trade-offs of including LLMs in your workflows are good for you, great!
Just don't delude yourself into thinking this new technology, this new tool, is free of trade-offs. Net positive for you? Ok if you think so go for it. I haven't found the inclusion of LLMs in my activities beyond replacing the broken state of traditional search worth it. That's it. I'm not against it, I just don't find myself getting net positive value from deferring to code generated by these tools, is all I am saying. Nothing more, nothing less than that simple statement. Not for me.
Maybe! But probably not, IMO. There have been a number of generations now that have been "addicted" to using digital spreadsheets rather than doing bookkeeping by hand.
I don't think this is in any way a disagreement between us. In just the same way that spreadsheet software did not kill demand for accountants, I do not believe AI will kill the demand for other professionals.
HN is often dominated by a Thing that is happening, during periods when there is some dominant Thing going on. In my time here, that has been many different Things over the years; social media, SaaS, Ruby on Rails, javascript frameworks, Rust, VR, self driving, crypto, and now AI (and this list is certainly missing a few). And during all those periods, lots of people have complained that the Thing of the moment is getting too much attention. But IMO, this kind of eye of sauron effect on a main Thing of a given time is mostly good. We're all here working out what we think of things, and it is useful to focus a lot of discussion on the new Thing that is developing and moving quickly.
I'm unconvinced that this is a pattern that is distinguished by the nation being reported on. Headlines vary. Both of your example headlines might appear at different times in the outlets I read.
I think this was just a stylistic choice and that this whole thread is a weird tangent.
My five year old who has never seen a Looney tunes cartoon thought it was hilarious. (It made me think he'd probably love actual Looney tunes cartoons!)
My 5 year old spent the whole time laughing. We have the expression "grow up!" that we tell kids. Sometimes I think we need the opposite expression for adults who are peeved when cartoons don't make them laugh.
To me, this seems to be an expectations issue. I thought it was going to suck. It's a live action and cartoon mix about looney tunes, itself a broad and very meh cultural artifact IMO. Compared to my expectations, it was funny and good. It seems that you had very high expectations for it to "elevate the genre", which ... I struggle to understand why.
Yeah but what is the point of "naked model" evaluations? It seems to me that the only interesting question is capability per time and cost (and conditioned on other important things, like privacy). It doesn't matter to me which part of that equation is being implemented by model weights and which part by other supporting software.
Currently if my hypothesis about frontier labs doing creative tricks between the model and the client is true (and the results seem to favour it so far) the benchmarks are giving us an artificially lowered results for open weights models.
I have yet to test opus/sonet via my proxy. If Qwen gets 10% better and Opus stays the same that suggests one if two things:
- either opus doesn't need it
- or it's already done behind the scenes.
In my view, the useful like for like comparison is to the entire system that people actually use. Nobody uses a "naked model", so what is the point of these benchmarks that use them in that way?
I think the benchmarks should be trying to use realistic harnesses for both proprietary and open weight models.
Harnesses can fairly easily be adapted to other models. Moving capabilities from one model to another is a more involved process requiring distillation, training, etc.
I don't quite understand your point. Why are you moving those capabilities from one model to another, or improving the built-in capabilities of a model, what is the goal? If having the capabilities in the model itself improves the overall capability, then using that better model with the same harness should achieve better results. Or if the capabilities are the same, but they've been moved from the harness into the model, then there should be some other detectable improvement, like to speed or cost.
I don't know about others, I can only speak for myself. But I do appreciate numbers for bare models, numbers for model + harness, numbers comparing different models in the same harness, and numbers comparing several models across several harnesses.
It's a lot of information to ingest, but it gives me some idea of which part of the system is doing which part of the work, how well different harnesses and models interoperate, and more insight into the part of the equation under my direct control as a software developer.
I agree that all else equal, the more comparisons the merrier. But IMO, there has been way too much focus on benchmarks for the bare models, when it seems to me that what actually matters is what you can do with the technology, and that is never limited to what a bare model can do.
reply