Question for people who choose not to use generative AI for ethical reasons: Do you make that choice despite accepting the growing evidence that it works (at least for some tasks, e.g. coding agents working on some kinds of software)? Or do you reject it because of the ethical problems *and* a belief that it doesn't actually work?
I'm thinking that principled rejection of generative AI might have to be the former kind, *despite* evidence that it works.
@matt How high is the bar for "works"? Code can compile and be a security nightmare, for example.
Relatedly, how long must it "work" for it to count? Do you have to be able to maintain the software?
Is the code still licensed properly? Did it "work" legally?
Did the chatbot give you mental illness which interferes with your ability to discern reality, and therefore to tell whether anything works?
Yep. “Works” needs to be nailed down pretty firmly here.
Case in point: LLMs as a tool for answering factual questions. They will sometimes get them right. They *cannot*, however, actually get there from first principles, the way a human can, and so they also get things wrong a lot. Hilariously, brutally incompetently, wrong. For example, I check in with the major LLMs with topics from my primary hobby (amateur astronomy) every so often, and they routinely botch things. ChatGPT 5.2 recently told me that a galaxy I wanted to look at from some place in New Zealand would be in the northern sky, so I should look in the *southern* sky to see it. And that’s not even the worst error I have seen it make. The logical conclusion: either LLMs are bizarrely incompetent at this one single topic, despite there being plenty of useful training data for it, or they are similarly incompetent at other topics and people have a hard time seeing it if they aren’t experts in the topic. I know which way Occam’s razor slices here.
It’s like that for most other use cases I’ve tried, so: regardless of the ethical issues, I not only don’t use LLMs, I actively avoid them.
The one “it works well” use case everyone brings up is software development. LLMs are *definitely* better at this than they were a year ago, but only in one narrow area: speed of writing (and, to an extent, testing) code that *nominally* meets specifications. But software development, especially as a profession, isn’t just about cranking out superficially correctly behaving code fast. I work as a software developer for a large company, and only about 1/4 of my time is actually writing code. A lot of it is making sure I know what my business actually *needs* from code, and a code-spewing machine doesn’t help with that at all. Additionally, I have run into a lot of AI-generated bug reports and task tickets where the AI-generated output was simply wrong, causing me to waste hours tracking down reality. LLMs are actually a clear net-negative value for me right now in my day job.
Matt Campbell
Nelson
David Nash