Post by @scottjenson@social.coop

There's a certain Dunning Kruger effect going on with AI. I've talked to other #UX designers and they find it liberating to be able to build prototypes no longer beholden to a programmer.

But that doesn't make them a good programmer, and it certainly doesn't erase the host of ethical and environmental issues behind this. I'm not saying "#LLMs are awesome."

What I'm trying to explain is WHY people are mistakenly thinking this is more powerful than it is in reality. This is where Dunning Kruger comes in: Person A being able to roughly do person B's job (poorly and without even knowing how bad they are) is at the very heart of why the #AI hype is so high.

#llms #ux #ai

1 Boost

0 Likes

Kyle Hughes

@kyle@mister.computer 2 hr. ago

@scottjenson Unfortunately this is not true anymore. Opus 5.5 is a better programmer than every human programmer I have ever worked with in my life, including myself, and I’m very good. A designer using it will have better outcomes than I could have delivered myself in 2023.

Michael Steeber

@michaelsteeber@mastodon.social 2 hr. ago

@kyle @scottjenson I still don’t think the latest models solve the problem of knowing what your requirements should be. Opus 5.5 can build you the wrong thing with technical excellence. Without an understanding of the architecture, it’s impossible to effectively steer it or provide meaningful feedback on a plan it makes. That’s probably fine for prototypes, but it’s where I still see major deficiencies in vibe-coded apps made by non-engineers.

Scott Jenson

@scottjenson@social.coop 46 min. ago

@michaelsteeber @kyle I do think programming appears to be the one domain where LLMs are clearly improving rapidly. And I can't argue with your point, as I don't adequate experience to judge it. But I have heard similar claims and it will probably get better over the next year. But I still think my point about doing other people's jobs holds.

Maybe just not as strongly with programming.

Michael Steeber

@michaelsteeber@mastodon.social 29 min. ago

@scottjenson @kyle The question I’m always asking myself is what is the “minimum viable knowledge” you must have to complete a task. I have no doubt the models will eventually eat the entire process up to that layer of abstraction, UX included. But there is just some point at which you must gather a little bit of understanding, even if it’s mediated by the model. When I buy a microwave dinner, I must at least glance at the cooking time on the box.

1 more reply

Kernel Bob :progress_pride:

@kbob@chaos.social 3 hr. ago

@scottjenson Do prototypes have to be good? If it holds together just well enough to express the designer's intent, isn't that success?

Maybe I don't understand what kind of prototypes you're talking about.

Scott Jenson

@scottjenson@social.coop 45 min. ago

@kbob My point really wasn't about prototypes. It was about the fact that I can use LLMs to be passable at something I know very little about. What's important is that I don't have the judgment to say if it's 'good enough' or not. I happened to approach it through UX and programming, but I think this applies to anything that people are doing with LLMs.