Hacker News new | past | comments | ask | show | jobs | submit
> I am extremely afraid that are on the verge of losing coherence on all technology.

I've worked for some large orgs in the past. One thing that surprised me, probably because I was so early in my career, was how there was only a few, sometimes only one, who knew how the system worked. When something went wrong and that person wasn't available we'd all gasp. Then, after swearing a lot, we'd gather round and roll up our sleeves. Debugger statements, coffee, then when it got late it was beer. Rinse and repeat. Line by line. I see no difference now. If an LLM writes something and I can't rely on the LLM to fix an issue it created guess what? I go line by line.

I guess it’s hard sometimes to discount experience and expertise down to “well, anyone can do it if they apply the right patterns”. Sometimes it takes a long time to understand all the bits and pieces to make any sort of reasonable trade off, or figure out why the trade offs there don’t make sense for the model. Sometimes solutions are imperfect but in a relative balance in terms of tradeoffs in one context, and when a piece of that context changes (goals, requirements, emphasized traits) it is tricky to fix without making things worse. Sometimes you just have a hard time pulling things apart.

LLMs seem to discourage people from building mental models a lot. I think those models are really important to keeping things working, and I think we’re better at it than AI because it can’t consider nearly as many contexts coherently.

When you don’t get to throw things at the wall and see what sticks with your customers or users data, you can’t yolo it, you have to hold a brain trust in how your things work.

I've definitely gotten away from modeling but thats a great point. When i started with LLM coding i would sketch a model and work with the agent to refine it. I don't do that now, just trust it. I suppose using it has made me lazy