If you think vulnerability research simply doesn't matter, you have a lot of company in that opinion. It's something Bruce Schneier used to argue, and Marcus Ranum, and a bunch of other people that used be on closed secret vulnerability-sharing mailing lists before Bugtraq blew those cliques up. These are very old arguments.
But if you do think vulnerability research matters, and you're trying to argue that frontier models aren't a seismic change for that discipline, you have almost no company. Vulnerability researchers are overwhelmingly leaning on automation to find vulnerabilities and, just as importantly, generate the tooling required to test hypotheses.
You can feel about that however you want to feel about it. I mostly don't care, except: you can watch people like this being negatively polarized back into the bad old days of the mid-1990s, content-free CERT advisories, and vendor-controlled "responsible disclosure" by a use case that frontier models unimpeachably excel at.
It’s amazing how half the developers on earth live on a completely different planet now. There are plenty of new challenges, sure but we are far past the point where we can have a debate about “is it useful?”
And yet we continue to do so. I understand the feeling of loss some people may be facing. And there are definitely some really bad practices - like nakedly spewing claudspeak at your colleagues instead of communicating. Or raising a PR you don’t understand. There are asymmetries we haven’t learned to navigate. But we aren’t returning to a world where it doesn’t dominate our discipline so it’s best to find opportunities.
The question most of us are asking isn’t “is it useful”, but “is it worth it”. Engaging with these tools involves no small amount of self-debasement and long term degradation of skills; do I want to sacrifice myself on the altar of productivity?
For me the answer is still absolutely resolutely “no”.
Because AI removes you from analytical thinking, and the brain is a muscle. The less you think, the less you are able to think.
Programming is inherently analytical and logic-driven. It’s a great brain exercise, even if the code has no value. When you use AI, you lose that exercise. What replaces it isn’t the same.
I mean, consider. I can solve a crossword, or I can ask AI to solve the crossword. AI will do it perfectly, and faster too. But which one is better for my brain? What was my goal with solving the crossword, anyway? On the surface it was to find a solution to the crossword. But, maybe, deeper down, the goal was to improve myself in some way.
People conflate the improvement of products with the improvement of self. Maybe it’s a result of our consumerist mindset. But the truth is the product can improve greatly, and you can be making it, while you yourself degrade.
We write better and better code, at a faster rate, but are we better programmers than before? Maybe, maybe not. I’m leaning maybe not.
It’s not new or unique. The assembly line is the same. Sure, I can build better furniture faster at the factory than by hand. But what is the goal here? Is it to make the best furniture, or to become the best carpenter?
If it’s to make the best furniture, the best product, then I’ve won. If it’s to become the best carpenter, then I’ve lost heavily.
The best carpenter obviously doesn't use any power tools. Are they allowed to use hand tools, or do they need to split the wood with their bare hands?
It's not clear to me why programming is inherently analytical and logic-driven but prompting and iterating with an AI is not. What about Applescript, where it's programming but in something closer to natural language?
The core of this seems to be that there are people who assume that users of AI turn their brains off the moment they open a Claude prompt. There are surely some people who are doing that, but there were also plenty of people who were shipping sloppy code before. There are far more people for whom this another tool in their toolbox.
If you really loved horseback riding, or you really loved breeding and training horses, or you were really good at it, I'm sure that cars fucked up your day, and that it sucked to see cars replace horses.
I'm sure there were plenty of people who said that cars were horrible and that we should stick with horses. I'm equally sure there have been C programmers looking derisively at Python programmers, Python programmers looking derisively at Node programmers, etc etc.
But there's a pretty big gap between the feeling of unhappiness that something you liked is no longer as in-demand / profitable / common / whatever, and saying that cars aren't useful and we're all worse off for letting our horse-riding skills degrade.
Normal people didn't ride horses everywhere. They weren't used like cars. I'm so sick of this analogy. People walked places. If it was too far to walk you used a horse drawn bus/tram. If you had to much to carry you used a horse drawn truck. Every single person wasn't riding an individual horse ten minutes each way to buy groceries every day like that do now with cars. Cars unquestionably kill drastically more people than horses ever did or even could. Tens of thousands a year in this country alone! The leading cause of death for children and young adults!
We're pretty off track given that ya'll are nitpicking the metaphor rather than the actual topic, but claiming that horses were never common doesn't really have any connection to the claim that per-mile-moved / per-human-carried, horses are also dangerous.
Never use analogies when arguing online. It only ever results in exactly this, people nitpicking the metaphor. You've essentially given them a way to "be correct", about something irrelevant.
This does not get to the core of it, LLMs are not equivalent to cars.
All of your analogies (in other posts) discount that:
- LLMs are not deterministic
- they do not produce high quality/expert level output.
- They are not easy to control consistently.
Once again, and i mean this in the nicest way possible. You do not sound like you understand how things operate at an expert level of engineering. I do not know of any senior engineer who thinks LLMs can consistently produce quality output.
You are way too invested in your argument and refusing to see other perspectives
You're not an outlier. The anti-AI stance is likely to be vocal online. They're quite incentivised to talk about it. Most programmers have adopted AI assisted programming (my own assertion) and most will feel like you and I, that it is useful. I know of a few programmers personally who hate it, but even they use it. There is no going back.
It is pretty obvious that when you come into a large incomprehensible codebase (which most are), AI will be able to reason about it more quickly and guide you through it.
I switched away from my work for a few minutes, but at the very moment I'm using AI to summarise some spaghetti code in a particular area in our codebase, so I can refactor it.
And that is my style of working at the moment. AI coding is useful, sometimes amazing, but sometimes irritating, and from experience I think just telling it to refactor this area would result in something I'm not happy with. So instead I'm getting a handle on it, I'll form the design I want, then tell it to do it. I have a feeling a lot of complaints are from people who won't bother to do that.
You are not an outlier. 40+ years of hobby and pro coding experience here and I absolutely love Claude Code. It enables me to dramatically increase code quality, performance, and features while still doing the fun parts and letting the LLM do the boring parts.
tptacek · · focus · HN ↗
But if you do think vulnerability research matters, and you're trying to argue that frontier models aren't a seismic change for that discipline, you have almost no company. Vulnerability researchers are overwhelmingly leaning on automation to find vulnerabilities and, just as importantly, generate the tooling required to test hypotheses.
You can feel about that however you want to feel about it. I mostly don't care, except: you can watch people like this being negatively polarized back into the bad old days of the mid-1990s, content-free CERT advisories, and vendor-controlled "responsible disclosure" by a use case that frontier models unimpeachably excel at.
jeremyjh · · focus · HN ↗
And yet we continue to do so. I understand the feeling of loss some people may be facing. And there are definitely some really bad practices - like nakedly spewing claudspeak at your colleagues instead of communicating. Or raising a PR you don’t understand. There are asymmetries we haven’t learned to navigate. But we aren’t returning to a world where it doesn’t dominate our discipline so it’s best to find opportunities.
vermilingua · · focus · HN ↗
For me the answer is still absolutely resolutely “no”.
akerl_ · · focus · HN ↗
Why is using AI tools self-debasing or degrading?
preg_match · · focus · HN ↗
Programming is inherently analytical and logic-driven. It’s a great brain exercise, even if the code has no value. When you use AI, you lose that exercise. What replaces it isn’t the same.
I mean, consider. I can solve a crossword, or I can ask AI to solve the crossword. AI will do it perfectly, and faster too. But which one is better for my brain? What was my goal with solving the crossword, anyway? On the surface it was to find a solution to the crossword. But, maybe, deeper down, the goal was to improve myself in some way.
People conflate the improvement of products with the improvement of self. Maybe it’s a result of our consumerist mindset. But the truth is the product can improve greatly, and you can be making it, while you yourself degrade.
We write better and better code, at a faster rate, but are we better programmers than before? Maybe, maybe not. I’m leaning maybe not.
It’s not new or unique. The assembly line is the same. Sure, I can build better furniture faster at the factory than by hand. But what is the goal here? Is it to make the best furniture, or to become the best carpenter?
If it’s to make the best furniture, the best product, then I’ve won. If it’s to become the best carpenter, then I’ve lost heavily.
akerl_ · · focus · HN ↗
It's not clear to me why programming is inherently analytical and logic-driven but prompting and iterating with an AI is not. What about Applescript, where it's programming but in something closer to natural language?
The core of this seems to be that there are people who assume that users of AI turn their brains off the moment they open a Claude prompt. There are surely some people who are doing that, but there were also plenty of people who were shipping sloppy code before. There are far more people for whom this another tool in their toolbox.
zahlman · · focus · HN ↗
Or, you know, to pass some time in an enjoyable way. To have fun.
timacles · · focus · HN ↗
almost every developer i know is fully aware, we are degrading every aspect of our skills.
The code writing, code reviewing, code understanding.
The more "time" passes and the more the code base grows, the more disconnected we become.
jeremyjh · · focus · HN ↗
akerl_ · · focus · HN ↗
If you really loved horseback riding, or you really loved breeding and training horses, or you were really good at it, I'm sure that cars fucked up your day, and that it sucked to see cars replace horses.
I'm sure there were plenty of people who said that cars were horrible and that we should stick with horses. I'm equally sure there have been C programmers looking derisively at Python programmers, Python programmers looking derisively at Node programmers, etc etc.
But there's a pretty big gap between the feeling of unhappiness that something you liked is no longer as in-demand / profitable / common / whatever, and saying that cars aren't useful and we're all worse off for letting our horse-riding skills degrade.
vermilingua · · focus · HN ↗
akerl_ · · focus · HN ↗
queenkjuul · · focus · HN ↗
akerl_ · · focus · HN ↗
queenkjuul · · focus · HN ↗
akerl_ · · focus · HN ↗
munksbeer · · focus · HN ↗
queenkjuul · · focus · HN ↗
akerl_ · · focus · HN ↗
timacles · · focus · HN ↗
All of your analogies (in other posts) discount that: - LLMs are not deterministic - they do not produce high quality/expert level output. - They are not easy to control consistently.
Once again, and i mean this in the nicest way possible. You do not sound like you understand how things operate at an expert level of engineering. I do not know of any senior engineer who thinks LLMs can consistently produce quality output.
You are way too invested in your argument and refusing to see other perspectives
akerl_ · · focus · HN ↗
akerl_ · · focus · HN ↗
AI has radically improved my ability to parse and understand new codebases, to validate my hypotheses, and to work in complex systems.
I wish I knew what I was doing differently than almost every developer.
munksbeer · · focus · HN ↗
It is pretty obvious that when you come into a large incomprehensible codebase (which most are), AI will be able to reason about it more quickly and guide you through it.
I switched away from my work for a few minutes, but at the very moment I'm using AI to summarise some spaghetti code in a particular area in our codebase, so I can refactor it.
And that is my style of working at the moment. AI coding is useful, sometimes amazing, but sometimes irritating, and from experience I think just telling it to refactor this area would result in something I'm not happy with. So instead I'm getting a handle on it, I'll form the design I want, then tell it to do it. I have a feeling a lot of complaints are from people who won't bother to do that.
deterministic · · focus · HN ↗
fragmede · · focus · HN ↗