More and more “AI”-driven plugins seem to be popping up in the Rhino ecosystem, promising fast and easy GH definitions, code, architectural plans, industrial designs, etc.
Currently, they still sit on top of existing software tools, but this might change rapidly.
I don’t think it’s too far-fetched to question whether Rhino — or any similar software — will even have a future.
I don’t doubt that there will always — even in the event of AGI — be a few passionate people who still want to work on a computer, perhaps as a hobby.
However, I imagine the herd might eventually thin out to the point where it is no longer financially (or otherwise) viable for companies to maintain their software.
Sure, an LLM could potentially take over development at some point — but what would be the point, right?
So my question to you all is: do you think there is a future for CAD and CG applications if a trained model could simply generate the desired result from a text prompt in the form of a ready-to-use file?
Personally, I think there may come a time when computer-based work becomes obsolete.
All these “tools” are just wrappers around one of the big LLMs–which are a security disaster you want nowhere near the data you make your living with–and have no path to being a viable business model. There’s a few of them spamming Reddit and they’re just a joke.
The training data for actual “AI CAD” is all locked up in our proprietary data, and I guess the big dogs in CAD are probably working on that, but there’s no evidence it’s here beyond hype from grifters on X.
Regardless, no work is not going go away, that’s not how any productivity improvement has ever worked except in the very short term. Human desires are literally infinite, there will never be enough robots for everything.
One thing is for certain, you can expect volatility in this regard; especially in subscription SaaS and expensive monolithic softwares as they fight to remain relevant and deliver on promises to shareholders and venture capital (neither of which Rhino answer to)
I just hope I have 10 years left….I have yet to see even a simple AI ad on that just predicts my next move much less reads the mind of an architect to figure out what they meant to draw as opposed to what they actually drew. I can’t begin to tell you how much the architectural drawings I have to use have went down hill over the past 35 years.
To loosely cide Roger Penrose: AI does not understand, it just computes.
It estimates/interpolates a solution for you out of huge datasets. This gives the immersion that it understands, but it does not. You can clearly see this when working on more complex problems, when there is little training data. Ai does not truly reason and therefore does not apply the learning to a problem where there is no training data. Instead it hallucinates and gives terrible wrong answers.
This essentially prevents the creation of new and tailor-made solutions, which is very common criteria for technological demanding solutions. But after all it will be helpful to develop more complex problems with less knowledge. So its clearly a useful technology. But it will not replace CAD nor any engineer. If that happens, than for other reasons.
Btw the worst aspect of this debate is about efficiency. Do you really need anything be generated faster? This only makes sense if the output is equal or better (quantity vs quality). And how many prompts do you need until you get exactly the result you want? I do by no means underestimate AI, I’m just tired of the “this changes everything” debate.
I work in industrial design and haven’t really seen an AI tool come close to generating a proper design. A design with wall thickness, bosses, ribs, draft, multiple components etc. So I’m not worried about the immediate future. I’m sure it’s coming though, but that would need to be coupled with sensory feedback systems so that changes can be accurately described using 3d printed parts etc. Or maybe this only happens properly when AI is housed in a robot body so that prototypes can be fitted / tested in the real world and feedback used to adjust the design.
Design doesn’t happen in a vacuum, and right now AI is still very much confined to a box.
for real. the people who claim “this changes everything” are the same people who’s day-to-day activities don’t contribute anything beyond what a computer could contribute.
agreed with your last paragraph. it’s usually 1000x faster to generate one precise thing that describes your idea than it is to design a tool that could spit out 1000 versions of your idea that are close-to-but-not-quite-right.
How can AI read my and my colleagues’ minds when we’re designing, say, a new ski boot, aircraft seat, or fork lift truck? Can it invent novel cushionings and mechanisms, make material experiments? Can it take all the real world material, machine, and production parameters into account? Can it build mock-ups for focus groups and user tests, and then change the preliminary 3D model accordingly?
Even the high-end AI industrial design exhibition by London studio Blond last year during the Salone del Mobile in Milan showed that none of the above is even remotely possible.
Somehow I feel that you don’t truly worked with people before.
AI is really good at making anything out of nothing, but AI is really bad at making adjustments out of something that already exists, without changing anything else.
I’ve had situations where 2 or 3 designers worked 3 to 4 weeks on one single curve to be able to achieve the shape that the client wanted, to the point were he was asking for milimeter ajustment on the shape and radious so it would “reflect right” the way he wanted.
IMO, the AI generated stuff will make it harder for designers to work because people will think that it is easily done and generated without knowing the true extend of details it takes to make things good looking and properly able to be fabricated and ask for slower prices or try to lower wages in the process.
Speaking from experience here. I had the absolute pleasure of working with a certain German car manufacturer on their “autopilot” system (NDA go brrr), and let’s just say…
Also I’ve read a few wrong post here, which I feel obliged to correct, let me clarify from a scientific point of view. The AI does understand, and that phenomenon is called grokking
The AI, initially learns by memory, then at a certain point it starts pruning activation patterns and leaves only the minium viable solution, at that point the AI has generalized beyond the dataset and has a real understanding of the task at issue, and we call that Grokking.
Grokking is a real phenomenon that separates memorizing, from understanding, and It’s all within the process of learning.
What we learnt is that if we keep running epochs after an AI has actually learnt the solutions, after some epochs the AI will actually learn and generalize beyond the given dataset.
This is called “Grokking” and It’s the phenomenon of understanding.
A fun fact for the nerds like me, the activation patterns after the pruning form geometrical patterns (beatiful ones really), and by running a fourier transformation you can actually decompose how the activation pattern solves the task at hand, literally understand what the AI underrstood.
This is just so you know, AI does not parrot, It learns and generalizes, the more it learns, the more well defined the grokking geometrical pattern is. You can run the experiment locally, I’ve done it on the sep256k curve for point addition, and I can confirm it factually groks.
Provided a quick trial to Diegodx, honest review. I won’t post the video of the platform because It’s a diegoDx design, he can upload it himself if he wishes :
Its not so clear that “grokking” is equal to understanding. At this point it is an edge case and practically not relevant. In my limited understanding its a phenomenon based on over-fitting for specialised AI. Over-fitting however can only occur if you have enough training data in the first place. And here I come back to my point. If you solve a problem with little related training data, there is likely no chance of grokking taking place. I mean I’m not ruling out anything for the future. At some point I’m expecting this change. However, I’m skeptical if we see this soon.
Btw, I work for over 13 years in the automotive industry and I do know how they cook (its not hot at all)…
You guys are all designers/coders etc… It’s easy with your experience to dismiss this stuff, but IMO that’s a risky thing to do – I think for you and for everyone else. It may not replace you or your job, but it will certainly change it – possibly quite radically.
There was this statement:
So how will the general public will react to products made with it? If it generates 1000 designs, 999 of which are crap but one sorta works, that one will indeed get produced. And many people will buy it, despite the lack of quality. Otherwise Alibaba, Temu, Shein, and most of what sells on Amazon wouldn’t exist.
And it’s only going to get better (or worse, depending on which end you’re looking from).
On a personal level, as AI can vibe code Python scripts for Rhino better than I can after nearly 20 years of learning/doing, it’s not really useful for me to continue with that. And I’m glad I never really learned to do rendering as AI can now do that for me as well (not that I really need it).
So I’m happy to go back to working on my old cars now, as AI can’t hold a wrench yet and by the time affordable AI-powered household robots will be able to, fossil fuel-powered vehicles will probably be illegal to possess and I’ll be in jail…
How can AI create a 3D model for a new fork lift truck, NX and Catia compatible for refinement and component suppliers, while taking the plethora of materials, production parameters, and tooling requirements into account?
To this day, in our/my research, we haven’t seen any AI application that is even remotely capable of the above, in a way, that would cut down product development time in a meaningful, real profit-generating, fashion?
I think hardly anyone dismisses it. Its more a matter of exaggeration and hype. I personally love to use AI for things which help me. But what worries me is when people put out their crystal balls and begin to predict when we are redundant.
The irony is that it likely learned from you since you are one of the people posting your scripts frequently for decades in this forum. That is a chicken egg problem. If you would not invested 20 years of learning/doing, there would be far less to learn from this place. But yes, maybe sooner or later they train themself in a generative way. But either way… it will never be as it used to. That’s for sure
do you have a video of the aurumflow plataform? you have to buy credits to use it, but here is no example of it working or the results you can get anywhere on the site.
