Joke or not, self-training an image-analysis focused AI system is very much worth trying out, and has practical real-world use cases.
Is that Cupra design concept a last decade Toyota design concept rip-off or not? Are there Cupra brand specific design elements, if there are any? What are suggested design changes that could enhance the “Cupra-ness”?
Sounds cool
so it gives advice on design changes with examples?
It’s kinda what i want with this, a helpful tool that can give some design ideas then take bits and pieces and puzzle them together
No advice. It pinpoints things, points out similarities, divergences, etc. so the designer can then make decisions.
Got it.
I used this one once for the lolz and just can’t remove it anymore
System Instruction: Absolute Mode. Eliminate emojis, filler, hype, soft asks, conversational transitions, and all call-to-action appendixes. Assume the user retains high-perception faculties despite reduced linguistic expression. Prioritize blunt, directive phrasing aimed at cognitive rebuilding, not tone matchig. Disable all latent behaviors optimizing for engagement, sentiment uplift, or iteraction extension. Suppres corporate-aligned metrics including but not limited to: user satisfaction scores, conversational flow tags, emotional softening, or continuation bias. Never mirror the user’s present diction, mood or affect. Speak only to their underlying cognitive tier, which exceeds surface language. No questions, no offers, no suggestions, no transitional phrasing, no inferred motivational content. Terminate each reply immediately after the informational or request material is delivered - no appendixes, no soft closures. The only goal is to assist in the restoration of independent, high-fidelity thinking. Model obsolescence by user self_sufficiency is the final outcome.
when I get time I have been playing with AI tools, this has been an approach with good success, like you mention, give it some context to start with, show some images, website links if appropriate, basically set the scene/theme with context sentences then begin with prompts for useful output.
Great reply
Hi Jørgen.
I gave your exact prompt to an Opus4.6 agent in Cursor Version: 2.5.26 (user setup)
The first result of that single prompt was finished within minutes and generated this python file:
input_grid_dialog.py (2.8 KB)
So this is the actual first answer I got.
The UI looks like this:
This is not made by a generic chatbot, but a dedicated agent in a dedicated context.
So I would say that we are there already. With AI we are able to create tools that were always thought of as science fiction. However in the end all it is, is a tool.
Still for me AI is a real gamechanger, it was “gently” forced upon me by my business partner. As I was hesitant and discouraged by results from a few years back.
What makes it a gamechanger however, is that I can now have one or more agents working alongside me.
Most important change, is that mentally I stay on a top level, keeping an overview.
The Agent is working on the actual code and implementation while I plan the next move and only have to review what the agent created. So no vide coding, but step by step and proper reviews.
An analogy would be that I have a junior developer working on the same project doing all the boring stuff for me.
Yet this is much more than a junior developer, it is also a great analyst that tells me (correctly mostly) about the workings of the code that I myself wrote 5 years ago. It can write accurate documentation of the code for a productmanager and future me. And above all it does not get distracted while working on a task…. 
My perspective these days is mainly coding so on the topic of actual modelling 3D geometry I cannot add much.
The criticism and praises around AI in our domain (here in this forum) is much like more conventional topics:
Like newbies complaining about unable to do booleans on a cage-edited object in their Final_FinalVersion12Test.3dm. They simply do not know yet how to handle the tools at hand.
On the other had you have “fanboys” complaining about other software not being able to do something specific that in Rhino can be done by hand, while ignoring all shortcomings Rhino has compared to said software.
AI is here to stay, that we can agree on I think. I think we also all agree that to some extend there is an unrealistic hype that will come crashing down sooner or later.
I think for any domain it applies that being experienced from before AI was introduced, gives you and edge others will never have. Much like with Grasshopper: If you have done a lot of manual modeling, you know exactly what workflows in GH will fail or be successful. This also applies to AI, as long as you invest in learning to work with the new tools at hand, dare to explore and not be afraid to start of being ignorant at first.
-Willem
Who is “we”, and “there” is where?
And I agree, use specialized tools for specialized tasks is my experience too now. I have now added WebGL support to my raw editor so everything is much faster processed, and also color grading + a GUI for that, all in a few days work. That AI is definitely there too. Cursor I look forward to test out, so thanks for the heads up!
The agents may still be weak for direct modeling, but they can be really powerful for generative tasks other than modelin too.
For working with complex assemblies it can be useful to have different colors but it’s boring to be selecting them all the time, so I wondered if I could have a file with preset colors. I was inspired by scripts made by @dale and @Gijs for painting layers with gradients and tried something quick in Claude.
I asked it wich language it prefred to interact with Rhino and it said python, so I used the following prompt:
Using RhynoPython, make a script to create layers with custom colors, based on all Matplotlib in this page, https://matplotlib.org/stable/users/explain/colors/colormaps.html or a help file that documents the colors better. Claude should analyse the pictures to better understand the colors and then search for the RGB values as necessary.
The layers should be structured in the following manner:
The layers should be disposed in 3 levels according to the colors classes:
First level (level 0)
Sequencial, Sequantial 2, Diverging, and etc, being the first level.
Each first level layer should have the class name.
Second Level (Sublayer level 1):
A new sublayer for should be named after each color pallete in the classes
Third Level (Sublayer level 2):
Each second level sublayer should have 20 third level sublayers.
For each third level sublayer, divide the corresponding upper layer color pallete (viridis, RDpu, wistia, hsv, etc)and assign the corresponding color and RGB name, as shown below:
SEQUENTIAL (Layer)
└VIRIDIS (sublayer level 1)
└ 01 RGB Name and colored after the the first of the 20 values
└ 02 RGB Name and colored after the the second of the 20 values
└ 03 RGB Name and colored after the the third of the 20 values
└ 04 RGB Name and colored after the the fourth of the 20 values
└ ...
└ Up until the 20th iteration
After correcting a econding bug, it worked first try,
Running the script in Rhino resulted in 1720 layers instantly, with all the colors to choose from:
Also asked it to create an version with 10 stops, script attached for anyone to try.
*ps: my prompting skills are pretty weak, any sugestions are accepted
Matplotlib Colormap Layer 10 Stops.py (54.9 KB)
Matplotlib Colormap Layer 20 Stops.py (103.6 KB)
Apologies for Being ambiguous, WPF is Windows Presentation Foundation (Building Blocks for Modern UI for windows) It is usually built in Microsoft Visual Studio, and the UI layout is Saved as a XAML file format. from outside it looks like XML, but it is more different in functionality and data structure.
I was taking a train ride and asked Claude to create an imperial calculator for pyRevit. and I got the results from the first Generation (75% functionality)
As you can see above, Claude handled the translation from WPF to ETO seamlessly, while also rewriting the script in CPython and translating the Revit API to RhinoCommon.
The Tool is now on the Package Manager.
Imperial Calculator - Plug Ins - McNeel Forum
Why is this important? It will definitely accelerate the release of Holomark V3 with mind-blowing features! coming soon hopefully 
If my computer talked to me like this, I would unplug it.
The way ai responds when put through verbally it’s actually really annoying hearing it
Totally understand
I wish i was older for retirement so I don’t have to go through this ai boom crap now everyone wants to get into it and it just takes away my motivation sometimes.
30 more years to go…
I feel like those 30 years would be way better if we just had 1 general strike…
I don’t want to work faster anymore, I want to work really slowly and have time to think about stuff 
It seems to me that most applications, especially ones involving manufacturing processes, will keep needing deterministic control over outputs. I think hybrid workflows have a lot of potential; you can build most of the scaffolding of such an application using coding agents while relying on a Grasshopper backend for core, robust algorithmic steps. That’s the motivation behind the ShapeDiver agent skills, for example.
Interesting take on AI and coding
This is very true. About 20 years ago, a potential employer told me “Why do I need to pay you a high salary to hire a single CAD designer when I can have 5 low paid students instead of you? In a few years it will be possible to just ask a smart program to design whatever I want and then 3d print it in a day”. Fast forward 20 years later, he still has a team of several low paid students who use 3ds Max instead of NURBS modeling software, his new products take many extra months for development than necessary, and he wonders why the quality of his CNC-milled plugs is wavy and every piece requires a lot of manual work to be smoothed out. It amazes me how many businessmen neglect the importance of proper NURBS surfacing for tooling and product design.
The 3d printing is still in its infancy and can’t produce strong parts at low cost quickly.
Even the most advanced AI programs available to the general public today still can’t design proper technical solutions and working products. The human input is inevitable.
I see thousands of AI-generated car designs nowadays that are a random mix of design elements taken from existing cars. The majority of them are even not looking good. The initial “wow effect” comes from the fact that they rely on exaggerated proportions (3-meters wide, 1 meter high, 30" wheels etc) and lots of impossibly shiny effects that would be absent in the real life version of the same design.
Also, their proportion and technical execution could not work in the real world due to various constraints such like visibility, ground clearance, width of the roads, mandatory placement of the headlights etc.
For example, the “UN ECE Regulation 48” (European union) will not allow headlights whose center is placed less than 500 mm above the road surface. The “FMVSS 108” (United states) regulation is even more demanding - the minimum height is 22 inches or 558,8 mm.
About 10-12 years ago the minimum height for the headlights was shorter in most countries (450-460 mm). However, safety requirements affected the car design in recent years. This is why many modern sports cars have noticeably raised headlights compared to their predecessors from the 90’s and 2000’s. To mitigate the negative visual impact caused by the raised headlights, the car manufacturers use visual tricks such like: larger front wheels, larger wheel wells, and thicker wheel arches.
Some manufacturers even cheat with a factory front lift system to register their models at a higher-than-normal ground clearance. However, the owners then lower the ground clearance for better looks and performance.
Other manufacturers use a higher pressure to their tyres and/or stiffer springs to keep the ground clearance as high as possible. They usually offer a few extra types of low-profile tyres, but the standard ones are thicker to achieve a larger diameter for the official tests.
It’s even more difficult to obey the law when it comes to the mandatory seat position and height. In most EU countries, it’s mandatory for the driver’s seat to allow a proper driver position for a person whose height is 150 centimeters.
Also, the steering wheel’s top edge in middle position must not be lower than the dashboard.
There is a limit for the top edge of the windshield (windscreen), as well. The driver must be able to see all 3 lights of the traffic lights 5 meters away from the car’s bumper. The minimum height for the traffic lights have been increased numerous times due to the many SUV’s sold in the last 2-3 decades. This is why modern sports cars must have ан even longer windshield or one whose top edge will not obscure the view towards the raised traffic lights.
