Definitely agree. Rear Window is essentially a perfect movie in all the ways I care about. Vertigo ultimately my favorite, but I wouldn't say it's perfect as much as I'd say it has much more ambition and ends up achieving more.

Luxray for sure! Give that pokemon something flashy, it needs it

I've had friends dismiss the water usage of AI and I constantly remind them that the real cost is in the local effects of data centers, not necessarily the global impact (but also that)

Damn. I grew up in Essex, I love to see these results!

I'm sure David Lynch would say it's a valid interpretation! Honestly it could be true. My perspective is a bit darker and I think Cooper essentially undid this ending.

Respectfully disagree! I remember my jaw being on the floor when I saw that in theaters. Nothing else was needed for me -- amazing movie.

I forget which comic said this, but I always think about: "I've never flown a helicopter, but if I see one in a tree, I feel safe in saying, 'he messed up'"

Resorting to directing an agent via natural language dozens or hundreds of times a day to edit a codebase also means that the risk for misunderstanding is compounded. IMO, until AI agents supersede human intelligence, it's just better to write your own code.

AI companies will likely try to compensate by continuing to train their models to better fill in the gaps by "knowing" general coding principles and philosophy. However, misunderstandings will always exist as a part of natural language, and I doubt this will ever be fully overcome.

One thing I learned very quickly when I tried these tools was that I am bad at being descriptive enough to have *anything*, human or otherwise, understand my specific vision for what I want to create. Some are good at this, but ultimately, a few paragraphs are never going to sum up a large project.

Of course, with AI, there's the implicit assumption that the tech will continue to improve, and therefore learning to prompt effectively is going to replace coding skills. It is very possible, but even as AI improves, will *we* get any better at telling it what to do?

I've seen some disastrous presentations from people who have "vibe-coded" apps and realized--during the presentation--that the app's logic fell apart completely upon any sort of inspection or testing. These are people so talented that they could have coded a comparable app in the same time.

Whereas coding bottom-up/from scratch ensures you know your code like the back of your hand, generating code and working top-down almost ensures you *don't* know your code. You will likely miss more bugs, and requested changes will be harder.

If you generate code via a prompt, you're likely going top-down: "generate me an app (or part of an app) in this framework that does these things." You then probably hit run, see what was generated, and start fixing issues--bugs, features you don't want, features that weren't added, etc.