What do I mean by some software devs are "ngmi"?
ghuntley.com
ghuntley.com
I've been railing against this since the get-go at my company. Initial responses were all scoffs, and I was basically out-grouped for my vocalisations on the subject. A year or so later, and people are realising that they actually can't use randomly inaccurate generators for work requiring reliable accuracy.
It just takes some people longer to learn, I guess. And some people need to experience real failure to learn.
And one of the biggest problems I see is that people who claim this kind of success with LLMs cannot reproduce the effort on their own. They do not learn any new solutions, because they didn't have to go through figuring it out themselves. So, good luck with that.
Of course, the possibility remains that it actually _doesn't_ take you 2 minutes to generate and verify code that takes 2 hours to write - that you simply have a bias from the dopamine hits of something else "solving" your problem for you, and have swept over the buggy outcomes with your rosy view. I've seen this a lot by LLM lovers - several people will come show me "convincing" demonstrations, and then _I_ have to find the bugs in the code, because they don't.
Thia is an extremely arrogant assumption. I have been programming for 20 years and your attitude is one I typically encounter in junior engineers. If you're not a junior engineer, I suggest you act accordingly and incorporate some humility.
> If you're talking about generating and verifying code in 2 minutes, I highly doubt that you're talking about more than 30 lines of code.
Another assumption, and another wrong one. I routinely generate hundreds of lines of code, generate tests, and review the code later once a unit of work is finished. An efficient workflow helps cut down wasted time.
> And one of the biggest problems I see is that people who claim this kind of success with LLMs cannot reproduce the effort on their own. They do not learn any new solutions, because they didn't have to go through figuring it out themselves. So, good luck with that.
Straw man argument, and a third display of considerable bias.
> Of course, the possibility remains that it actually _doesn't_ take you 2 minutes to generate and verify code that takes 2 hours to write - that you simply have a bias from the dopamine hits of something else "solving" your problem for you, and have swept over the buggy outcomes with your rosy view.
A fourth display of bias, increasingly incoherent and vitriolic.
Is this always how you speak with people? Do you sit down with a list of biases about them and deliver them one-by-one while presenting straw man arguments and avoiding discussion of the actual premise?
Perhaps if someone doesn't typically tackle novel solutions and takes pride in writing the same thing over and over, an agentic code generation tool wouldn't make sense to them. I don't know. But I don't think you're equipped for this discussion. You're getting increasingly defensive and emotional. I'm done, feel free to respond however you'd like.
-jrflowers”