> 1. The people on HN tend to think they're very clever, but they're often not as clever as they think they are. That has a lot of wide-ranging implications, from classic "engineer's disease" thinking to stuff a lot dumber.
Related: the Rationalists display this very well (though it's certainly not unique to them). The jump from "I think my beliefs through carefully, therefore I am reasonably rational" (likely already false, or at best true in a limited domain) to "I am rational, therefore my beliefs are well-thought through" is actually very small in the human mind, despite the latter being the opposite of rationality.
Also, the existence of Gell-Mann amnesia should be taught to everyone!
> 3. I think there's a certain amount of defensiveness from AI enthusiasts.
I've encountered this as well. I think this is also a huge part of why you see so many people loudly exclaiming "every programmer uses LLMs now" despite that not being even close to true. If you criticize LLMs, it (socially, to them) counts as criticism of their users as well. As well, bosses are probably the most excited about LLMs because they hope to fire us all soon, so it can be pointlessly self-destructive to speak negatively about LLMs in a professional context right now.
That, plus AI boosters tend to be incredibly annoying about it, bringing it up at every possible opportunity, changing every technical discussion to be about them, suggesting them for the stupidest use-cases, loudly proclaiming nonsensical productivity boosts (perhaps you feel 5x as productive because you're forcing other people to spend their time fixing your shitty slop code?) etc. So as someone who does not use LLMs I avoid talking to them.
I just think this speaks more to people observing different things and taking that as ground truth. I notice the people who are critical of these tools because like the above post and yours, they are relentlessly critical without ceding basic things like people are excited about these tools because they are useful. I have 0 doubts some people are seeing 5x productivity boosts because of the stuff that is regularly posted on HN - stuff that used to take months taking days.
I'm not a booster, I barely use these tools but what I notice is the two apparent "sides" that people have shifted into and really I don't think that' necessary because these are just tools that have adnvantages and disadvantages.
> I have 0 doubts some people are seeing 5x productivity boosts because of the stuff that is regularly posted on HN - stuff that used to take months taking days.
I will cede this: it indeed seems like you can throw together a shitty half-working prototype very quickly using LLMs. There's been a deluge of half-working prototypes, but years in and where are all these projects now? No study has concluded anything near a 5x improvement.
Besides, my actual opinion is that even if these actually gave a large productivity improvement, it's wouldn't be worth it. The negatives are simply too large:
- the concentration of power they are pushing for;
- the environmental effects of training;
- the disinformation they spread at scale (which in practice is their primary use-case);
- the adverse effect on a shared reality from the fact that they cannot reliably answer questions, yet are presented as doing so;
- the adverse effect on a shared reality from their sycophancy (do the rich surrounded by yes-men seem in touch with reality?);
- their anti-social nature, as they discourage engaging with other people in favour of solipsism masked by a poor simulacrum of personhood;
- their weakening of labour power because of the managerial class' belief that they will soon replace workers (whether or not they actually can is not particularly relevant);
- the large-scale plagiarism they enable;
- lastly, their weakening of already incredibly lax standards of engineering from our field.
The promise of more software does not even come close to outweighing all of that.
It's just from my personal experience when I rarely use Claude, I can get it to produce things in minutes that would take me hours. They're not shitty, I can review what's been generated and it's better than I would have done. Similarly with the UIs we build here, they would take days or weeks and we can put them together in a few hours.
I can't speak to you other issues but these are here to stay and I think the only realistic way forward is to learn how to use them.
> I can't speak to you other issues but these are here to stay and I think the only realistic way forward is to learn how to use them.
People only say this about things that they want. Ignoring the consequences of one's actions is not a neutral position. Societies are perfectly capable of banning things (yeah, you can probably sneak a local model through a hypothetical law, but it would end the industrial scale problems).
There isn't any evidence. It's pretty much all anecdotal. I can only speak from my experience.
> People only say this about things that they want. Ignoring the consequences of one's actions is not a neutral position. Societies are perfectly capable of banning things (yeah, you can probably sneak a local model through a hypothetical law, but it would end the industrial scale problems).
We're not going to ban these models, though. Which means we have to figure out how best to live with them.
> A recent study of actual coding practices across hundreds of firms finds that human code review forms a significant “bottleneck” for the overall efficiency of AI coding tools, resulting in “little evidence that firms increase software output or reduce employment” by using them.
> When developers are allowed to use AI tools, they take 19% longer to complete issues—a significant slowdown that goes against developer beliefs and expert forecasts. This gap between perception and reality is striking: developers expected AI to speed them up by 24%, and even after experiencing the slowdown, they still believed AI had sped them up by 20%.
LLM boosters always cry foul with this one because newer models exist (see next item), but the fact is this is evidence that programmers are remarkably bad at estimating their own productivity.
> For the subset of the original developers who participated in the later study, we now estimate a speedup of -18% with a confidence interval between -38% and +9%. Among newly-recruited developers the estimated speedup is -4%, with a confidence interval between -15% and +9%.
With the rather large caveat:
> Unfortunately, given participant feedback and surveys, we believe that the data from our new experiment gives us an unreliable signal of the current productivity effect of AI tools. The primary reason is that we have observed a significant increase in developers choosing not to participate in the study because they do not wish to work without AI, which likely biases downwards our estimate of AI-assisted speedup. We additionally believe there have been selection effects due to a lower pay rate (we reduced the pay from $150/hr to $50/hr), and that our measurements of time-spent on each task are unreliable for the fraction of developers who use multiple AI agents concurrently.
I guess there's a way to use them productively. I only use them rarely which means I use them when I know they'll speed me up and the tasks take a few minutes. I literally could not do the tasks in less than an hour or two, the same with UIs that get generated here, it's just impossible to be faster than the models.
> How could you possibly be confident of that?
I just don't see any indication that we're anywhere close to that. It would take a rather big shift in governance.
Related: the Rationalists display this very well (though it's certainly not unique to them). The jump from "I think my beliefs through carefully, therefore I am reasonably rational" (likely already false, or at best true in a limited domain) to "I am rational, therefore my beliefs are well-thought through" is actually very small in the human mind, despite the latter being the opposite of rationality.
Also, the existence of Gell-Mann amnesia should be taught to everyone!
> 3. I think there's a certain amount of defensiveness from AI enthusiasts.
I've encountered this as well. I think this is also a huge part of why you see so many people loudly exclaiming "every programmer uses LLMs now" despite that not being even close to true. If you criticize LLMs, it (socially, to them) counts as criticism of their users as well. As well, bosses are probably the most excited about LLMs because they hope to fire us all soon, so it can be pointlessly self-destructive to speak negatively about LLMs in a professional context right now.
That, plus AI boosters tend to be incredibly annoying about it, bringing it up at every possible opportunity, changing every technical discussion to be about them, suggesting them for the stupidest use-cases, loudly proclaiming nonsensical productivity boosts (perhaps you feel 5x as productive because you're forcing other people to spend their time fixing your shitty slop code?) etc. So as someone who does not use LLMs I avoid talking to them.