AI Thoughts & Sketch-Up

I’ve been thinking about AI, and even writing articles about it, a lot lately, like a lot of people. I know a lot of people are afraid it will take away jobs and make architects and engineers obsolete, but I think this is hyped up fear. Yes, AI can produce pretty pictures, even semi-workable architecture if prompted right and iteratively. But buildings can’t be kinda, sorta, right. They must be precisely engineered, structurally sound. I know. I spent 14 months with another skilled SU user on this forum working for 1,000s of hours with >14,000 messages back and forth on this forum, skype calls, and emails, until finally coming up with this:

5th Floor Quarter Floor Layout

Do that with prompts! Yes, I know you can prompt AI to “design an arched building over the lower East River of NYC to house and provide workspace for >20,000 people with 1 million square feet of commercial space…” etc. etc. It won’t work. It’ll take so many modifications that the prompter would tear his hair out and it would never be right anyway. I doubt it could even do this:

I’m not an architect or an engineer, but Dami Lee is and she did a video about AI a while back that’s still relevant:
She and her team were given a tight competition, since AI is so much faster, but looking closer at its models and there are glaring problems, like paths to nowhere, even impossible structures that wouldn’t work in real life, or would be expensive side extensions for no purpose. I think as they would have drilled down even more, AI would become even less useful. The video is not a deep enough dive, IMO.

AI can do a lot. It can even do beauty. But it can’t read minds. It borrows from everywhere and can’t originate a single thing that isn’t derivative - this is a much bigger handicap than most people realize because they themselves don’t do original work. When you DO original work, you very quickly realize how limited AI really is.
The fancy parametric architecture that is all the rave today is NOT Zaha Hadid, it’s just in her style. None of these constructions will work without a lot of modification the old fashioned architecture and engineering way. And SU, whether AI-extended or not, will be here through any advancement in AI too.


:rofl: AI is not the only one who does this.


Already confirmed by major A&E firm’s then Chief Exec, and internationally recognized architect too - with a letter. And developers would have come on board (besides the small one who did already). The problem is permission and half the site is owned by Con Ed which recently decided to keep their pier for future repurposing as a clean energy hub.

I think posting these articles could make for a more interesting post than promoting your project.

There is nothing interesting here related to AI, despite all your hours spent that is anything other than ‘I worked hard on this and I think AI isn’t capable of my brilliance’.


Here they are, all rounded up from 2023:

I hesitated to post them because none of them have anything to do with what Sketch-Up does or could do, but since you asked…

It’s not that I’m so brilliant, but that my work is the closest at hand and easiest to post in detail. And pictures are worth a thousand words and are posted here in response to all the AI pictures so easily generated by the new art of Prompting, yet fall short.
I did talk about Dami Lee and Zaha Haddid too, but in the first case, Lee structured her own competition with AI tools in a way that disadvantaged the strengths of her own team, and in the latter case, it’s just a case of AI imitating the late Haddid, without capturing her originality nor her level of workable details (which are not available to me to post either). The Devil, as they say, is in the details. I just don’t see how one gets to a workable building without ultimately taking over from AI’s 30,000ft level imaginings. And that means more work, not less, for architects, engineers, and modelers like myself.


There are new tools all the time. I’m going to see if GetFloorPlan can improve my layout: First, I wrote to them to see what kind of files they process with AI. It’s all offline and sent back to the user for a $25-$45 fee for 2D/3D images. I’ll post a new render here if there is one.