You’re very observant! (Maybe just checking if it is actually a human)
1 Like
a massive issue right now is that while you have the expertise to see the mistakes, a crapload of users don’t.
and it’s like bugs in code, if users don’t tell chatgpt it’s making mistakes, then there are no mistakes to fix. and users believe it to be true.
4 Likes
I partially agree with what you have said but usually if you run the code the AI generates and it gives wonky results you can pretty quickly determine that it isn’t working correctly or something is slightly off.
For example yesterday I wan’t to create a very concise piece of code to generate a sphere at the origin. I already know how to do this, so yes, I can debug the code quickly and determine if it looks to be in error. But when I ran the code ChatGPT generated it did not produce a sphere, it tried but it failed. It was interesting how its attempt was more or less correct but it somehow missed one simple thing and therefore it would not produce the desired geometry.
Even if I did not understand any Ruby I would see that the code was not giving me the desired result so therefore I could easily report back to ChatGPT that its code was no good.
oh yeah, sure, with code there is a test in the end.
your code either works or it doesn’t.
but looking at knowledge in general, when people use chatgpt as a search engine, they don’t know the answer. I’ve had people show me chatGPT answers in a ah-HA ! moment except it was an incorrect answer.
the same way someone could feed a photo of something to an AI and get a 3d model. if they don’t know how to model, nor how to design the specific thing, they don’t know how good or bad it is.
at least until they start using the part and it blows up in their face I guess 
it is quite impressive indeed. yesterday I gave this photo to model a theatre chair : (I used meshy, free plan)
I ended up with this obj (that took approx 2 hours to import in sketchup )
not bad eh ?
except it has 230 000 edges and 150 000 faces.
I ended up modelling it myself, as replacing autocad blocks by such a model just kills any computer.
It still is amazing how smart it can be, being able to add details that are not even shown on the picture, structural elements etc. and i’m pretty sure it will very soon be able to produce optimized models
4 Likes
And yet it is missing key features visible in the original photograph e.g. the seat pivot stops. In which case you could have just taken a few measurements and created a low poly representation in less time?
2 Likes
That’s what I find: nothing from AI is ever “off the shelf” ready (code, CAD, narrative, etc.) and it takes just as much or more time to clean it up as it does to just do it myself. For code and narrative, it might present a good outline, but I still worry I’m going to take it for granted and leave an error in there.
Don’t get me wrong, photogrammetry is great and a proven scientific application for sure (I spend a lot of my work day referencing ASPRS standards), but models like what you show are best suited for reprojecting the texture to something modeled by an actual user with a heartbeat.
3 Likes
Sort of off topic, just an AI ramble
One of the issues I see with AI is kinda like something that shows up due to the film industry, and social media etc. I pick on the film industry as perhaps the earliest thing that anyone alive can relate to in this sense.
We watch films/TV etc and we learn from it, or should I say we learn what it tells us, things that are not necessarily true but slowly become seen as true. All too often ‘facts’ are created by repetition rather than reality. Films/TV/Books/Word of mouth/Old wives tales etc show us things the way they want, or often the way they think it happens but it’s not that way at all and it become the norm.
Again I use film as they were the social media of their time, and I use something I am intimately familiar with. Glass, simple glass and stained glass more specifically.
How many times have you seen a burglar cut a circular hole in a pane of glass and simply pull it out? Totally impossible! but most people think it is true. There is a very specific technique to making a hole in a piece of glass and you’ll never end up with both pieces intact.
How many times have you seen someone thrown through a stained glass window and it shatters into a beautiful shower of rainbow particles? Never going to happen! Most likely they would thump against it and bend it and break the glass a bit then fall back on the ground on the inside feeling very sorry for themselves.
My rambling point being, the more AI learns from its mistake the more mistakes I think it will make. Same as the more people learn from google/meta/insta the stupider they become because they are learning from the algorithm that is designed to feed them what they want to hear. How else did the world get the orange blob.
Don’t get me wrong, I’m sure parts of AI will be incredible, but currently humanity seems to be prone to idiocracy as a goal.
2 Likes
This was something I was reading about last year actually. The worry that once AI is producing the bulk of images and text on the internet that it will no longer have human made content to feed off so will just eat itself until its producing recycled and unintelligible nonsense.
It also makes me think about the the Twitter bot they tried to get to talk like a person but it got fed so much racist and misogynistic stuff they had to shut it down. This could be covertly done to other AI models to influence there output without the creators even knowing.
1 Like
Only today in the old school televisual news broadcast, no doubt fake news, they had a story about one of our larger banks launching an anti phone scam ai system. Believe it or not it is an AI bot that answers calls from scammers and keeps them talking, record so far is 55minutes, hilarious and useful for now. But what happens when they start talking to each other.
When they start talking to each other you know someone has managed to create a problem and solve it and get paid twice.
I know a guy whose startup is making an AI calling system for businesses like barbershops (his original target). right now it answers the phone in order to give info or book an appointment so the barber doesn’t have to stop to pick up. the step he’s working on is having the AI call people to reconfirm the appointment and eventually call people to notify them of an opening earlier than planned.
give it a few month / years, and an AI will call you, and your ai will answer the call.
And one will be the client ai and the other the architect ai and they will develop a forum where they can talk to each other about those dam bios that keep building houses manually without plans…
5 Likes
I watched that video - definitely super interesting stuff.
Also read a bunch of the comments - I know the CGI space is being heavily impacted by technology like this.
I think the gap right now is that AI is great at spitting out a model based on a prompt, but it’s really not great at knowing what it created and helping you make changes.
It’s the same problem I currently have with SketchUp diffusion and other AI rendering tools - I don’t want you to CHANGE my model when you render it - I want to see an accurate representation of the design that I’ve created, which is usually based on a discussion or many discussions with the users of the space.
I’m interested to see if the models are actually capable of starting to understand what they’re creating - that seems like the real barrier between AI really taking over a lot of the design process.
2 Likes
I don’t think we are making better buildings with the technology we have - in fact, the soul and poetry is exactly what’s missing!
My daughter’s friend does social media for the Empire State Building and he took us on a back-of-the-house tour. The ESB was built in 13 months during the Great Depression from hand drafted drawings with art deco detailing and beautiful craftsmanship - you feel the imprint of the minds and hands that made it one of a kind. That building exudes the love and pride that went into creating the world tallest building during a time of great hardship - a shared sense of purpose.
Two new high tech buildings with competing observation decks have recently gone up in NYC. They are cool and feature state of the art tech but have nothing on the ESB and I don’t think they will inspire the imaginations of movie makers, musicians and artists - in fact, they look like they could be in any other modern city and I cant readily remember their names or what they look like - just lots of glass. The ESB is the only one of the three that is literally irreplaceable.
I hope that it’s a pendulum and as a reaction to the lack of work in the profession in it’s current form - young professionals will use their creativity and the amazing tools we have to reconnect with the art of building and reimagine how architecture is practiced and the ways it can celebrate humanity.
5 Likes
Your view of the Empire State Building is a bit misguided.
The project was designed to be as economical as possible, particularly the facade design was the least expensive at the time. It was designed with strictly financial rules (and not love) with narrow lobbies and spaces no more than 30 feet up to the windows to create the most profitable offices possible (the best criteria for leasing at the time). There’s also an architectural history in New York with the industrial advances in electric lighting. The Chrysler Building is much more of an architectural masterpiece than the Empire State Building.
You may find these explanations about the Empire State Building in the book “Form follows finance” by Carol Willis.
Hey Alan4 – just wondering why your public profile is hidden…?
Of course you are well within your rights to do that – just asking…
Honestly, I’m just too busy to do it properly.
If I put even one line in there, it’ll trigger questions, replies, follow-ups… and I barely have time to answer my own thoughts
Well, maybe you should take some time off and just do that.
When will we see your new 3D engine? Will it compete with SketchUp?
2 Likes
Very interesting work ! Perhaps I’ll respond more broadly on the dedicated discussion thread?
In summary, I now have reliable tests, I’ve learned and discovered a lot, and I’m looking ahead.
The engine isn’t a competitor to SketchUp’s, as I’m working on a vector-based approach… but maybe not. In terms of performance, I’m not sure of anything at the moment.