AI influence on CAD-CAM

[QSIMDO]

You knowledgeable folks see any effects of AI, pro or con, on design and machining?

[RB211]

Fusion360 wants me to spend money on cloud credits to use their "generative design" feature. Basically produces organic shapes that are optimized for weight vs strength and such designs look impossible or stupid to design for subtractive machining. Best for metal printing.

[Dan Dubeau]

I still use cad from 2006 and CAM from 2010 at my job.....Safe to say it'll be a while before I see the effects in my day to day lol.

[QSIMDO]

What about design piracy?
Do you foresee problems along that line?

[RB211]

What about design piracy?
Do you foresee problems along that line?

Far too much effort is placed on protecting designs. As soon as it's released to the wild, everyone that wants to will copy it.

Everyone needs to get used to the cloud, it's never going back to how it was.

[dalee100]

At some point, AI will become a thing for engineering/design/machining.

But the "Boss" or sales won't be the ones using it. They can't be bothered to make the effort to learn it. So they will continue to hire people who can do those jobs. You need to know how to manipulate and direct the AI by asking the "right questions" to make it "work".

What it will do is make those people who actually do the engineering/design/machining and can add the skills of using AI effectively a lot more efficient at those jobs. Which management will see as a good reason to reduce the number of employees needed to accomplish the same amount of work. This may or not be a good thing - only the future can tell.

As it is right now, AI is no danger to people who can think creatively. AI can't currently do nuanced interpretation or have that flash of insight into some all new and never done for. Nor will it be able to for a very long time I suspect.

[CarlByrns]

It seems it's currently impossible to remove bad or wrong information bc of the way AI gathers info, so if a material's physical properties are recorded wrong anywhere, they're wrong everywhere.

[Rich Carlstedt]

It seems it's currently impossible to remove bad or wrong information bc of the way AI gathers info, so if a material's physical properties are recorded wrong anywhere, they're wrong everywhere.

And that my friends is the greatest challenge we face !!!!!
And "Who" determines those bits ?

Thanks Carl

Rich

[DrMike]

It seems it's currently impossible to remove bad or wrong information bc of the way AI gathers info, so if a material's physical properties are recorded wrong anywhere, they're wrong everywhere.

This is exactly why AI cannot ever live up to the hype. It's happening as we speak (maybe not in CAM, yet), everywhere on the Internet.

Websites are currently (actually have been for quite a while) updated using AI, and AI is always trolling websites for "new" and "confirming" content. Garbage content that is fed into AI that will be replicated (and "confirmed") in future content that will be found and replicated (and "conformed") by AI, ad infinitum.

AI will kill itself, since it believes everything it finds as "fact."

[thin-woodsman]

If you want a good laugh, talk ChatGPT through creating some OpenSCAD code for you, then render its output.

That said, the headline-grabbing stuff usually represents the *entertaining* use of AI, not its most effective use. These are stories written for people who either love AI or hate it, and just serve to confirm their opinions. Like so much if what's out there.

If you check out the model hub on HuggingFace (yeah, makes me think of Alien too), there are quite a few different approaches to designing an AI model, and they tend to get very specialized (image segmentation, image description, image generation, text summarization, *meeting* text summarization, etc), and these are trained on different datasets to produce different mdodel versions. For example, there's a Code Generation model trained on 7 billion inputs (whatever they define an input as, for that model): one version is trained on code from multiple popular languages, another is trained entirely on Python code. Python code produced by the second version should be superior (I say "should" as I have been to busy to write the code to verify this), as specialization of the training set generally makes for more reliable output. Which is why anything trained on the Internet is going to produce garbage

Back to the OP's question, there is certainly potential there. The CAD models are, or at least could be, text that is generated by an LLM, so that you say something like "The spec is for a plate of 1/4" thick 304 stainless steel, 3"x4", corners rounded with a 3/8" radius, and there is a 10-32 tapped through-hole in each corner, and a 3/4" hole in the center of the plate" and it would generate aCAD model of the part. We're probably about a year away from that, or less - I've been writing Langchain agents which interact with FreeCAD, Inkscape, and Blender (they all have Python APIs, as does most of the AI world right now) to experiment with this, though of course the models themselves will need a lot of help in producing usable output. It's ain't magic; nuthin' is. The trick with these generative LLMs is to make them the piece that talks to the user, and given them plenty of tools to use which produce the actual answers that the AI then provides to the user.

[DR]

I have not fooled with ChatGPT. Can it generate Arduino code for such tasks as running a stepper motor? If so, that could be useful for me.

BTW, One of my neighbors teaches at a private school with tuition around $50K per year. They have decided to allow student use of ChatGPT because it's the future and students need to learn to use and understand it. That school is also expanding due to classroom shortage and increasing applicants for enrollment.

[RB211]

I have not fooled with ChatGPT. Can it generate Arduino code for such tasks as running a stepper motor? If so, that could be useful for me.

BTW, One of my neighbors teaches at a private school with tuition around $50K per year. They have decided to allow student use of ChatGPT because it's the future and students need to learn to use and understand it. That school is also expanding due to classroom shortage and increasing applicants for enrollment.

Not only Arduino, but any microcontroller platform. Tried it with ST, and native ATmega C code.

[QSIMDO]

I have not fooled with ChatGPT. Can it generate Arduino code for such tasks as running a stepper motor? If so, that could be useful for me.

BTW, One of my neighbors teaches at a private school with tuition around $50K per year. They have decided to allow student use of ChatGPT because it's the future and students need to learn to use and understand it. That school is also expanding due to classroom shortage and increasing applicants for enrollment.

Trouble enough thinking on their own now as it is!

[Tobias-B]

Garbage in, Garbage out.

I think the Simulated Intelligence we're seeing now is just another intermediate step...

...I'm not sure we have the faintest clue about the direction we're walking.

t

[old mart]

As long as you don't supply the mill with any gunpowder, the AI won't be able to make a gun to dispose of it's owner.