This is the first TH-cam video I've seen about using AI to read drawings and create Gcode. It is early in its development but as things get better, it will be amazing in what it can do. I can't wait to see it happen.
I think if your goal is to improve productivity while reducing labor, the better path would be to have the AI learn to use CAM software, rather than trying to write the gcode directly. This can leverage the decades of toolpath algorithm development, toolpath verification, post processors for different machines, etc., while allowing to user to more readily tweak the results.
But if the AI is left to decide on its own what a toolpath should or shouldn't be, maybe AI can come up with better toolpath algorithms on its own. Interacting with existing software will limit its ability to that of the software it is using. This leaves CAD CAM as nothing more than an interface to interact with the AI. All very interesting as to where it ends up!
@@CamInstructor This is what Deepmind did with thier Go AI that beat the world champoin. They just made it play against itself a couple billion times. It made a move nobody had ever seen before and most though It was a mistake. It ended up being a genuis level move. It is crazy to think that ChatGPT was only released 10 months ago, and the difference in performace between then and now. I think it will be learning CAD in no time.
You can do this and it will likely work better like you say, but you will need to put in about $10m to develop a first version of this. Big tech are not going to build this and give it away for $20 a month like ChatGPT4.
@@CamInstructorIt probably doesn't come up with a better toolpath, because there's no intelligence in it. It just predicts the next token in the text, which includes g-code paths that are generated by existing cam software
1. Give it some G-Code, ask if it can improve the code 2. Get it to comment up code 3. Can it optimise code? 4. If you want to repeat the code multiple times, for example, because you can hold 10 items, and want the same ops on all 10. Can it structure the code to make it efficient? For example, minimise tool change time by spotting all holes first, one tool change, then drill all holes. 5. I think it has to go back to the CAD/CAM software, where you need to the AI.
Using plug-ins, wolfram alpha can be used to do the math, thus it's much more likely to get it right. You can also use the 'advanced data analysis' mode where gpt can write python code to calculate the math for the positions.
When using the plugin version of ChatGPT, you no longer have access to image upload. So, I gave it text instructions for the bolt circle...it only got 1 hole right.
@@CamInstructor Oh I see, hope they add support for combining the modes soon. Thanks for testing! Seems like 'a picture is worth a thousand words' is true here too.
We were looking at creating an AI bot with UI Path, but just for monitoring new doc releases. Helping us ensure the shop had the most up to date Engineering. I work in aircraft manufacturing, and they have recently approved 3D printed ti clips for use in commercial aircraft construction. We also use G code in automated track drilling and CNC cut and milled shims to exactly match contour. This is game changing but will be slow to get adopted due to FAA part certifications. Takes forever and also requires a recertification of the larger assembly and aircraft. Red tape for years.
I still don't have access to upload images. BTW, In your custom instructions for "How would you like ChatGPT to respond?" add something like this: I expect a list for required tools and tools locations, G-Code compatible with a Hoss CNC Milling machine. I expect you to plan the complete process before creating any G-Code. Keep in mind offsets, tool diameter, pocket milling (climb vs conventional), spot drilling, speeds and feed, After generation the G-Code evaluate it and rate the G-Code on a scale 1-100. If there are any calculation errors correct them.
Make sure you aren't using it in one of the other modes, like plugins. Needs to be in the Default mode. Click on GPT-4 at the top of a new chat, click on Default.
I'm self-taught in CAD/CAM and being able to understand G-Code in general. That might sound more positive than it really is. I learned enough to be able to use my CNC mill and lathe with good results. When I retired, I sold off my machines thinking I would have no further use for them. Now I'm wishing I'd never sold them because being retired has given me way more free time to work on my own projects and ideas, now that I no longer have my CNC equipment to work with. I've been thinking of buying another smaller CNC mill to keep in my home shop. The problem is that I've forgotten alot of what I learned about CNC machining and G-Code. It would be awesome to have some sort of AI assistant to help get me back on track with what I used to know about CNC machining. How far away are we from having an AI type of assistant with the ability to minimize the shortcomings you pointed out in your video? Also, with advances in this AI technology specifically meant as a tool for aiding in CNC machining, I suppose we can expect for this to come with a price tag that will go higher and higher before it becomes common place and the price starts to come down again. What kind of prices do you expect for the inevitable better quality CNC oriented AI systems?
As far as I know 3-d printers don't have g-codes, you upload a cad model in the right format, and the printer prints it, I may be wrong and that stuff might exist, but usually no coding is required in 3d printers unlike cnc.
Used it for 3d printing, My machine wasn't working properly and always messed up on the first few layers. I was at a loss until I asked AI to fix the parameters for my machine and it has worked better ever since and with the added benefit of faster cleaner prints. The default settings used to work flawlessly when the machine was new but AI through them away. Also support structures come out looking like trees now using less filament. I haven't asked it to make the models for me yet. I'm not completely ready for that because I like having creative leisure for this side of things, but I may consider it for monotonous or overwhelming models.
For CNC I think ai would be helpful- if it can get 90% of the code correct, that means a person doesn't have to work as hard getting everything written manually. Just have to correct the mistakes and it's done
We could use another programmer. There’s work aplenty and I struggle to keep up. I think AI will really streamline some of the chores of my job and leave me more time to do the stuff I like, actually making parts.
It could, the question is when. There is still a long ways to go, it all depends on how fast this can learn and if there is a limit on what it can produce accurately.
I'd love to see more what this might look like in something like FreeCAD so we can describe a modification to a part in text and with some back and forth make a change to it.
There are so many instances of G code, it's pretty much machine specific, this is why we have post processors. Still.. it seems it's getting there and will be pretty cool when I can just describe what I want and AI produces the machine ready code
The more that AI learns about the correct processes or rules, the better it will apply them. AI CNC coding should look amazing in a years time from now if it had a lot of input from users, with corrections recommended. Time will tell, but it's looking promising.
Nice to see that Ai evolving now in Productivity , ive seen many Videos now , where Ai even create Nature like Materials or better say the look like it ,but by increase strength and also flexibility so even the employes was astound of it. Also improving Material coast, workload and workflow in efficiency. Great Future we can watch in his childshoes getting better and better and this also in a fraction of time compare to other evolving technologies often takes years to expand.
Excellent video thank you. What about making molds with complex organic shapes? How far away is being able to make a high quality scan that gets translated simultaneously into G-code?
Can Chat GPT4 use bearings (Deg, Min, Sec & quadant 1=NE 2=SE 3=SW 4=NW) with distances in feet and decimal feet to create coordinates, lines in Autocad?
Pop quiz: at 2:28 after it drills the 2 holes on the left the drill moves diagonally from bottom left to top right. Would it be faster if drilled the bottom right hole before the top right hole?
Cutting across the diagonal would take a tiny bit longer. During some testing it did the angle move sometimes but not others. I'm sure if you specify to take the most optimal path, it would.
@@CamInstructor Since it's a rapid move and the hole pattern is square, it does not matter. Remember that the machine is using 2 axes to interpolate that diagonal move so no matter what, at least 1 axis will have to move 2.25" the other axis moving at the same time is 'free". In fact, even on a rectangular hole pattern it doesn't matter as long as the tool only traverses the longest direction once.
I've seen ai makes a lot of mistakes when it comes to inticate math like fourier and other calculus. It runs into issues with order of operations, if you remind it of the rules it adjusts and adapt to the correct answer, but it is not as easy as saying this is what I want, this is the problem. But that is speeds up time to give a 100% correct answer, that is why it is so powerful.
This is really interesting. I use chat gpt for software development and it can be useful but a lot of the time, it just makes stuff up. Sometimes it provides some insight and sometimes produces good code but you have to be very careful and never trust what it says. One thing to use for software and cad simulation but I’d need to be really sure before I used this on a machine that cuts something
I checked the engineering designs made by ai looks pretty cool, its gets a new perspective on things, a chair is a good example, is a chair really the best way to seat? What an ai can do to improve some common designs so we don't need to try many variants of the same things...Ai seems good on patter recognitions, but still need a human to tell them if its correct or not.
The variability of different machines will be an issue for AI. Itll have to have specific details of your particular machines in your shop. It wont inherently be universal. As the video showed itll basically give you a rough draft that youll have to fine tune.
This is crazy I’m sure a company like DMG Mori is monitoring all of their machines in real time and collecting vast amounts of data (similar to JGB equipment) to create AI algos.
I'd like to see it use a solid model, maybe even a native solidworks file that might give it the extra info it needs. I've said for 15 years I wanted a giant "do it" button. Looks like it might be in sight now.
I don't use it yet, but thank you to keep us updated on the progress. I think mastercam allready as artificial intelligence when using automatic regions in dynamic milling. I guess its a matter of time for GPT to master all the 2D toolpaths and starts to learn the 3d stuff, and verify the its code himself in the simulator.
Why not build a language model of all the cnc programs you have run on a given machine and the manual for the machine. Can the bot understand 3d models?
One of the first things I did with chat got was ask it for some code, it was basically correct but would not have ran on a machine. I of course praised its response and that it should not change a thing. No need to train my replacement. 😉
I love the future. Now I'm waiting for a open source programm like airbus/apworks used to calculate that organic skeleton form of the light rider motorcycle. PS: the earlier version of chat gpt was already pretty good with html-code.
It has a long ways to go! And also needs better instructions from operator....T1 = spotdrill (not T3-as at 3:09......this can confuse the logic of the GPT, unless specifically instructed to not follow tool numbers in ascending order-which is common practice. For chatGPT to understand that, it would add another logical hurdle and to overcome it would require additional instructions)....Maybe one day in the future...🙂
I think their automation is more the quoting end, though I'm sure they have in house automations setup...none of which you would have access to...and thats where ai like this helps everyone.
@@CamInstructor I've been told by people in the industry that their 3axis cam and 2 axis turning is 90% automated, generally just reviewed by humans or humans program specific requirements. But it wasn't from prototolabs employees, might be total BS.
I applaud you for exploring new ways of accomplishing tasks. The problem with AI is there is no I (intelligence). There is no creativity. No digital computer that by definition can only make decisions with on/off will never have intelligence. The intelligence is provided by the programmer. AI is nothing more than a super fast search engine, that goes through a huge database, and then outputs information based on rules defined by..... a programmer.
The computer that was trained to play GO beat the best players in the world using strategies no one had used before. It taught itself how to play, programmers did not. Same goes for chess, computers are better than humans. AI might be behind humans in many aspects, but the gap is closing quickly.
You should have uploaded a screenshoot of the incorrect result back to chatgpt to see if it could correct its error based on the image. Alternatively you could have told ChatGPT that the bolt hole pattern looks like a square or diamond shape and to try again. It will probably be a little while longer before chatgpt can one-shot things like this.
I did some back and forth trying to get it to correct the hole locations, even tried as you suggested, uploading an image.The image I showed it had the red circles in the correct location and black points were its locations, I then asked it if it looked correct. It was able to see it was not correct and attempted to fix. Still had a few holes off but interesting none the less.
@@CamInstructor At least ChatGPT tried, even though it still could not get it right. I am a chatgpt plus subscriber, but still have not gotten access the new image input feature. I was probably one of the earliest subscribers to chatgpt plus, but it always seems like I am one of the last ones to get access. Very frustrating. I have serveral things I would like to try as well.
Honestly it just needs a feedback loop- for instance just sending the generated image of the generated locations vs the expected and see if it can spot it owns errors
ChatGPT is my coding partner when I do C# or C. I do not use it for G-code. Same as normal code, it still goes off in strange tangents sometimes and forget some options available and you have to prompt debug it a lot so I feel for G-code the convectional tools are still more headache free than AI. In a generation or two down the road, when it have millions of tokens is will preform much better on all fronts. And it is really bad at making STL files too. lol. It will try but it will fail. haha.
We are well on our way to a world where not only can AI make the design for parts and whole machines, but actually design them from the ground up and improve on (evolve) the design over time. This means that there could come a day when machines entirely designed by AI are so advanced - after potentially thousands of iterations - that us humans would not even be able to understand how they work and would not be able to repair them when they fail.
Tip. If chat gpt makes a mistake and you know what the mistake is, tell it to fix its mistake explaining where it went wrong, ie order of operations and it will correct itself.
Most of my experience has been with 3D printers and CNC engravers, but I'm pretty sure most CAD software has gcode outputs... Edit: Okay, so this is a bit different, but if you wanted to be able to auto convert drawings into g-code that shouldn't be a particularly challenging thing to accomplish
I think it is worth mentioning that you have to turn on the the feature to allow uploads, currently in Beta, otherwise this video is confusing to a new user.
@@CamInstructor I'm not sure then, I had to enable this and I have the 'plus' subscription service, otherwise I don't have this functionality. I'm using Google as the default browser, I don't know if that has anything to do with it. I had chatgpt4 for quite a while before signing up for the plus version so that maybe also be a factor also. Not sure what to say.
Wasn't really meant to be a head to head test for whats faster. More just to see what it can do. If its able to do it, it will eventually be the faster method, just a matter of when.
I have an interest in GCode, GpT4 and manufacturing. My major issue is I don't have a brain. I am not an engineer. I want to work with an engineer(s), so ideas can be produced from ideas I have. I have been waiting for the news you have brought, but I am intriqued with type of errors of precision being made by the high level Machine Language program. As you have demostrated most problems can be debugged, and straightened out for correctness. This is obviously what must be done. More fascinating is the LLMs that are open source. As you know, NASA has decided they are going into real estate on the Moon. Nasa has an itch to put a residential area in outer space. That is fine, but they are going to need some help. And based on your video, you have shown vital information that will allow machines to be built on the Moon and not on Earth. Basically, as one scale's up, it would be a good thing not to write home saying we need this and that, when such a thing can be built by material that exists on the lunar surface. Very good job on the video production . Well done.
I do not use AI in my daily engineering work (matlab/simulink in my case). But I feel soon I will, coz they are really going to advance it in a year or so.
Most CAD/CAM packages on the market can already automatically generate tool paths. FWIW, I wrote programs during the late 80's that ran on Z-80 CP/M systems that could read DXF files and apply MY logic to automatically write programs for Bridgeport milling machines and Strippit-Houdaille turret punch CNC's. The advantage of my code was that it always worked vs. the garbage that modern AI generates.
@@LyonScapeStudios AI is capable of producing very low quality code, whether that is for CNC machining or writing software applications. I've been testing AI's ability to write software applications. While the output would seem amazing to someone who had just graduated college with an advanced degree in computer science, it is readily apparent to a skilled professional that even the best of code generated by AI is riddled with defects, including design patterns which will limit performance, stability, and scalability. I see AI for what it is - a con game that can be used to deceive the gullible.
@@gaiustacitus4242 thats your problem. You are assuming it will write code because thats what you do. It will not, it will just utilize it. Like an ape, it will use the sticks and stones around it to complete tasks. And preliminary testing has been stellar.
@@LyonScapeStudios The entire purpose of the video was to demonstrate the capability of AI to automatically program CNC machine tools. Likewise, I've watched other videos and read articles touting the ability of AI to write software applications for computers and mobile devices. I'm not "assuming it will write code." I'm testing the claims, and so far I've only been able to dispel them. AI is not supposed to be taking existing "sticks and stones" and use them for assembly of its output. The claims for AI tout its ability to LEARN how to generate NOVEL works. As demonstrated in this video, AI's efforts to generate an equally spaced hole pattern used the correct formulae but it failed to properly apply it when calculating the X/Y coordinates. AI also has a problem with using conventional milling for CNC cutter paths, when even a neophyte knows that climb milling is required for CNC operations (even if ignorant of the reason for doing so). You look at AI and see it through rose colored glasses. I look at AI and laugh because while it is Artificial it most certainly has no Intelligence.
@@gaiustacitus4242 chat gpt is the tip of this iceberg. If you bothered to do any deep diving you would know that. There are many manufacturers worldwide that have completely automated programming solutions and they are very good. Please do some more reading before you take such a strong stance.
Slightly surprised that ChatGPT doesn't 'get' trig: Sine and cosine range between [-1,1], the expressions ...,1.5 + sin(36 deg) should have read 1.5 x sin(angle), multiplying the radius by the value of the circular functions, not adding them.
That's not what it did wrong. You can't see it in this video but the radius of the bolt circle is 1 so it's a unit circle with a center point of (1.5,1.5). The expression 1.5 + sin(36 deg) is correct and that scares me even more. It should have been correct.
@@danarrington2224 I asked it to explain the formula it used for that location, hoping to see where it went wrong. While it was explaining, it realized it gave an incorrect value and then apologized lol. Asked it to double check all the locations and let me know how many it had incorrect...it found a few more but even after some more back and forth, it continued with at least 2 incorrect locations. Not sure why.
@@danarrington2224 Ah - I mistook a "3" I saw earlier (found the frame at 0:51, for the square drill pattern) for the later example's diameter Bolt circle of 2" dia on 3" stock (at 4:07). Got it. That IS odd.
i think the issue here is... people expect chatgpt to instantly know what u want based on random words u input which is ridiculous cause honestly talking to a human if u tell him concise words and sentences to the point most humans dont understand... so expect less from a machine that learns and does 100% exactly what u ask it... so if u ask something wrong... it will give u something wrong... i got the best results from chatgpt by explaining in detail everything i wanted and some possible issues also... so it can think of them and take them in calculations... and 99% of the time worked fine... there is that 1% when as u said... mid talking with it... forgets stuff... or just disregards it... cause if u ask it... based on the input above or first thing i asked.. it knows... so its all user input fault honestly... its a matter of understanding how to perfectly work with it... i find best results is give it small fractions of a big thing u working on... dont give it a full bulk load... lets say u can split a cnc program in 10 steps... ask it to code that entire project but ask it in 10 small parts drill some holes, do some chamfers etc etc... and at the end u cut off ur programming time by more then 50%. u can also tell it to skipp math lessons, just to give u the results as u are not interested in a explanation.
It takes too long to specify all parameters. You have to check the result in a CAD and correct it with AI perhaps multiple times until you get the satisfying result. AI is good in learning but it cannot learn or predict what the final product will look like which exist in your head only..
Must be the trig throwing it off. I was doing some TDAmeritrade Thinkorswim/Thinkscript programming. The program gave errors at first, but once I added screenshots, it fixed them right away. Could likely be fixed by an app call to WolphramAlpa.
This is crazy. I had asked chatgpt to write some code for a haas mill, and it said it did not have a model for the code. That was a year ago. I even asked Ai to add the code for a robot to unload and load a new part. Hold on tight the skill gap is closing
Could anyone here do sin of 36degrees in their head without a calculator? If you allow the Language Model (not math model) to use a calculator, like you did, it will get the calculation correct.
Indeed, the model is able to use knowledge of prior calculations of the sin of 39 degrees, the issue is that ChatGPT uses a non zero "temperature", this value can vary from around 0.5 to 1 with 0 being almost deterministic recall and 2 being whacko off the wall choices. With that in mind you can see how the models recall when exposed to this randomness (artificial spontaneity, lets call it) can go a little off. I'm a community champion over at community.openai.com and you're super welcome to drop by and I can walk you though maybe creating some API code where you can specify temperatures and other fine tuning of parameters, and even get you started with the Playground :D Always exciting to see new users and CNC was very interesting to watch.
1) Use the "Advanced Data Analysis" it is better for coding and can actually write and run Python code including drawing and simulation 2) Ask it to do "step-by-step" it backtracks and reread what it has done and often time it improves to an extent; 3) Invest in AI companies funds, in a few years there will be nothing that you can do on a computer that AI will not do better, faster and cheaper, it will only get smarter from here and, at a certain point there will be no way to make money other than by investing and some niche jobs. The same laws of physics applies to us and machines and given enough development, the machines will eventually take over.
Image recognition is brand new. Yes, lots of machines were/are able to work with dxfs to create simple 2D paths, this is much different and a glimpse of where things are going.
Interesting but the way it operates so far is not conducive to such exact logic work. You might think that it fails on simple tasks but it's not using a calculator like you are, it's just doing the calculations on 'intuition' of math. Try answering your own prompt as if you were chatGPT and write it sequentially without ever going back and editing and reordering and without using a calculator. No notes on paper, just write it out immediately in your head. So it's not quite ready for being used in the real world but it's glimmers of amazing capability.
Oh this is just great...now they can start designing their T-800 Bodies. Thanks for the end of the world !!! But other than the beginning of the destructiion of Mankind, this was a Great video !
Don't you need more information, HSS or Carbide drill bit, # flutes & type of endmill , coolant? Good start but your programmer or machinist is left with fixing the program, so I rather have a skilled person do it in a CAM software package, and who is doing the machine set-up? This might help students get a basic understand of G-code and using AI but when I teach I have to get students to understand G-code (adjust program code if needed on shop floor) and the different software used in machining plus all the tooling, fixtures, and the touch offs, etc. Like the video its a good lesson on AI advancing in manufacturing processes such as CNC machining.
Ai is just complex matrix math to design algorithms to produce data from a dataset. What jobs ai creates or takes away is a natural part of human development. For all of recorded human history, horses and blacksmiths were an essential part of human society that was completely replaced in two decades. But now we have more mechanics and engineers and business people and all sorts of jobs for people because of the car industry. There are more people working in the jobs cars created than cars ever replaced. Same with computers. There are more than double if not triple the human population on this planet than when horses and blacksmiths were phased out. Ai will make more complex and fulfilling jobs more accessible to those who might have been limited to less fulfilling jobs in more limited capacities. AI will not replace many jobs, but people using ai will. Learn these things now or you will be replaced. Any industry and job will require you to stay current to stay in that job or industry. AI is just the catalyst this time.
0:31 you have to upload drawing if want ai programmed do it no intelligence. now your top secret drawing is with them. sure they want you upload like apple want your files on cloud
The fail with Gcode and math itself is well, it might seem underwhelming. But consider this - people who wrote the code for ChatGPTs AI - most probably have NEVER HEARD OF GCode!!!! Let alone understand it. Probably same is true with trig stuff.
Not good enough. I don't want to make a drawing in the first place. I want to take an old part from my car, put it in a 3d scanner that is integrated into the CNC cabinet and have the CNC machine spit out a new part BASED on this that is in accordance with the specs and NOT a direct copy of the used part. So basically it must identify the part, look up the specs online, and just make the part WITHOUT me worrying about looking over GCODE or other stuff.
So you are applauding ChatGTP's ability to translate a completed design into CNC instructions. In Computer Science, we call this translation. Fortunately, the process requires no intelligence or creativity. It has been studied since the 1950s. The creativity and intelligence are required to generate the design in the first place. (I will admit that most "designs" are minor modifications of previous designs. Creativity is very very very rare. Too bad there are no Expert Systems to help with this process.)
You have this completely backwards. AI is already able to do design, look at AutoDesks Generative Design. The machining is the hard part that it cannot do, yet.
This is the first TH-cam video I've seen about using AI to read drawings and create Gcode. It is early in its development but as things get better, it will be amazing in what it can do. I can't wait to see it happen.
I think if your goal is to improve productivity while reducing labor, the better path would be to have the AI learn to use CAM software, rather than trying to write the gcode directly. This can leverage the decades of toolpath algorithm development, toolpath verification, post processors for different machines, etc., while allowing to user to more readily tweak the results.
But if the AI is left to decide on its own what a toolpath should or shouldn't be, maybe AI can come up with better toolpath algorithms on its own. Interacting with existing software will limit its ability to that of the software it is using. This leaves CAD CAM as nothing more than an interface to interact with the AI.
All very interesting as to where it ends up!
@@CamInstructor This is what Deepmind did with thier Go AI that beat the world champoin. They just made it play against itself a couple billion times. It made a move nobody had ever seen before and most though It was a mistake. It ended up being a genuis level move. It is crazy to think that ChatGPT was only released 10 months ago, and the difference in performace between then and now. I think it will be learning CAD in no time.
@@silentbob1236Released, yes, created years ago though.
You can do this and it will likely work better like you say, but you will need to put in about $10m to develop a first version of this. Big tech are not going to build this and give it away for $20 a month like ChatGPT4.
@@CamInstructorIt probably doesn't come up with a better toolpath, because there's no intelligence in it. It just predicts the next token in the text, which includes g-code paths that are generated by existing cam software
1. Give it some G-Code, ask if it can improve the code
2. Get it to comment up code
3. Can it optimise code?
4. If you want to repeat the code multiple times, for example, because you can hold 10 items, and want the same ops on all 10. Can it structure the code to make it efficient? For example, minimise tool change time by spotting all holes first, one tool change, then drill all holes.
5. I think it has to go back to the CAD/CAM software, where you need to the AI.
Using plug-ins, wolfram alpha can be used to do the math, thus it's much more likely to get it right.
You can also use the 'advanced data analysis' mode where gpt can write python code to calculate the math for the positions.
When using the plugin version of ChatGPT, you no longer have access to image upload. So, I gave it text instructions for the bolt circle...it only got 1 hole right.
@@CamInstructor Oh I see, hope they add support for combining the modes soon. Thanks for testing!
Seems like 'a picture is worth a thousand words' is true here too.
@@CamInstructorthis is now changed, all modes can be combined! So image analysis and then code interpreter and wolfram alpha at the same time.
I did some quick calcs. I think it did sin(36) in radians not degrees. That’s why they were all wonky in rotation angle.
Nice…
We were looking at creating an AI bot with UI Path, but just for monitoring new doc releases. Helping us ensure the shop had the most up to date Engineering.
I work in aircraft manufacturing, and they have recently approved 3D printed ti clips for use in commercial aircraft construction.
We also use G code in automated track drilling and CNC cut and milled shims to exactly match contour.
This is game changing but will be slow to get adopted due to FAA part certifications. Takes forever and also requires a recertification of the larger assembly and aircraft. Red tape for years.
Here we go people. Teaching it to manufacture is the best idea ever!
I still don't have access to upload images. BTW, In your custom instructions for "How would you like ChatGPT to respond?" add something like this: I expect a list for required tools and tools locations, G-Code compatible with a Hoss CNC Milling machine. I expect you to plan the complete process before creating any G-Code. Keep in mind offsets, tool diameter, pocket milling (climb vs conventional), spot drilling, speeds and feed, After generation the G-Code evaluate it and rate the G-Code on a scale 1-100. If there are any calculation errors correct them.
You need to be on the subscription, GPT+
@@CamInstructorthat’s a gpt4 vision feature. I’m a plus user and still don’t have it. There are a lot of us
Make sure you aren't using it in one of the other modes, like plugins. Needs to be in the Default mode. Click on GPT-4 at the top of a new chat, click on Default.
I'm self-taught in CAD/CAM and being able to understand G-Code in general. That might sound more positive than it really is. I learned enough to be able to use my CNC mill and lathe with good results.
When I retired, I sold off my machines thinking I would have no further use for them. Now I'm wishing I'd never sold them because being retired has given me way more free time to work on my own projects and ideas, now that I no longer have my CNC equipment to work with.
I've been thinking of buying another smaller CNC mill to keep in my home shop. The problem is that I've forgotten alot of what I learned about CNC machining and G-Code. It would be awesome to have some sort of AI assistant to help get me back on track with what I used to know about CNC machining.
How far away are we from having an AI type of assistant with the ability to minimize the shortcomings you pointed out in your video? Also, with advances in this AI technology specifically meant as a tool for aiding in CNC machining, I suppose we can expect for this to come with a price tag that will go higher and higher before it becomes common place and the price starts to come down again.
What kind of prices do you expect for the inevitable better quality CNC oriented AI systems?
Please could you talk more about AI and Gcode in 3D printers?
Also, have you had any success asking GPT to code in Openscad?
As far as I know 3-d printers don't have g-codes, you upload a cad model in the right format, and the printer prints it, I may be wrong and that stuff might exist, but usually no coding is required in 3d printers unlike cnc.
Used it for 3d printing, My machine wasn't working properly and always messed up on the first few layers. I was at a loss until I asked AI to fix the parameters for my machine and it has worked better ever since and with the added benefit of faster cleaner prints. The default settings used to work flawlessly when the machine was new but AI through them away. Also support structures come out looking like trees now using less filament. I haven't asked it to make the models for me yet. I'm not completely ready for that because I like having creative leisure for this side of things, but I may consider it for monotonous or overwhelming models.
I use Chat GPT4 everyday for IT Support. Its very good for that purpose.
For CNC I think ai would be helpful- if it can get 90% of the code correct, that means a person doesn't have to work as hard getting everything written manually. Just have to correct the mistakes and it's done
Sure, but who writes Gcode manually in any kind of professional setting these days?
We all use CAM softwares :)
@@Robinlarsson83 Fanuc based old cnc lathe users do xD
I’m excited for AI in CAM programs. I use Solidcam at work and it would be nice if AI handled clunky stuff like updating the tool library.
Totally agree!
Till one morning the boss call you and thank you for the time you spend at his company.
Hope that you will still be exited that day.
We could use another programmer. There’s work aplenty and I struggle to keep up. I think AI will really streamline some of the chores of my job and leave me more time to do the stuff I like, actually making parts.
Thanks, right up my alley. These are great discoveries!
wow. Cool. I've often hand tuned cam software output anyway - could this eliminate cam software? That will be a big market disrupter.
It could, the question is when. There is still a long ways to go, it all depends on how fast this can learn and if there is a limit on what it can produce accurately.
I'd love to see more what this might look like in something like FreeCAD so we can describe a modification to a part in text and with some back and forth make a change to it.
There are so many instances of G code, it's pretty much machine specific, this is why we have post processors. Still.. it seems it's getting there and will be pretty cool when I can just describe what I want and AI produces the machine ready code
The more that AI learns about the correct processes or rules, the better it will apply them. AI CNC coding should look amazing in a years time from now if it had a lot of input from users, with corrections recommended. Time will tell, but it's looking promising.
Agreed!
Nice to see that Ai evolving now in Productivity , ive seen many Videos now , where Ai even create Nature like Materials or better say the look like it ,but by increase strength and also flexibility so even the employes was astound of it.
Also improving Material coast, workload and workflow in efficiency.
Great Future we can watch in his childshoes getting better and better and this also in a fraction of time compare to other evolving technologies often takes years to expand.
Excellent video thank you. What about making molds with complex organic shapes? How far away is being able to make a high quality scan that gets translated simultaneously into G-code?
This wild! Can it do lathe/turned parts?
Haven't tried lathe yet. I'll give it a go and report back!
Can Chat GPT4 use bearings (Deg, Min, Sec & quadant 1=NE 2=SE 3=SW 4=NW) with distances in feet and decimal feet to create coordinates, lines in Autocad?
Pop quiz: at 2:28 after it drills the 2 holes on the left the drill moves diagonally from bottom left to top right. Would it be faster if drilled the bottom right hole before the top right hole?
Cutting across the diagonal would take a tiny bit longer. During some testing it did the angle move sometimes but not others. I'm sure if you specify to take the most optimal path, it would.
@@CamInstructor Since it's a rapid move and the hole pattern is square, it does not matter. Remember that the machine is using 2 axes to interpolate that diagonal move
so no matter what, at least 1 axis will have to move 2.25" the other axis moving at the same time is 'free". In fact, even on a rectangular hole pattern it doesn't matter as long
as the tool only traverses the longest direction once.
@@danarrington2224 Oh good catch! Ya, even though its travelling further, its taking the same time when moving 45deg.
There have been simple programs that have created CNC G code from CAD DXF files for decades generally called Post Processors that didn't make errors.
This is not the same.
I've seen ai makes a lot of mistakes when it comes to inticate math like fourier and other calculus. It runs into issues with order of operations, if you remind it of the rules it adjusts and adapt to the correct answer, but it is not as easy as saying this is what I want, this is the problem. But that is speeds up time to give a 100% correct answer, that is why it is so powerful.
Does the Wolfram plugin help with the calculus? (currently requires gpt-4)
This is really interesting. I use chat gpt for software development and it can be useful but a lot of the time, it just makes stuff up. Sometimes it provides some insight and sometimes produces good code but you have to be very careful and never trust what it says. One thing to use for software and cad simulation but I’d need to be really sure before I used this on a machine that cuts something
Yes, it still has a way to go before its ready for prime time on the machines.
I checked the engineering designs made by ai looks pretty cool, its gets a new perspective on things, a chair is a good example, is a chair really the best way to seat? What an ai can do to improve some common designs so we don't need to try many variants of the same things...Ai seems good on patter recognitions, but still need a human to tell them if its correct or not.
The variability of different machines will be an issue for AI. Itll have to have specific details of your particular machines in your shop. It wont inherently be universal. As the video showed itll basically give you a rough draft that youll have to fine tune.
This is crazy I’m sure a company like DMG Mori is monitoring all of their machines in real time and collecting vast amounts of data (similar to JGB equipment) to create AI algos.
Has there been some change? I run ChatGPT 4 and it does not have an import image function.
Top left corner, make sure you are set to GPT4 and not GPT3.5. When in 4, you should see a paper clip icon in the input box
I'd like to see it use a solid model, maybe even a native solidworks file that might give it the extra info it needs. I've said for 15 years I wanted a giant "do it" button. Looks like it might be in sight now.
I don't use it yet, but thank you to keep us updated on the progress. I think mastercam allready as artificial intelligence when using automatic regions in dynamic milling. I guess its a matter of time for GPT to master all the 2D toolpaths and starts to learn the 3d stuff, and verify the its code himself in the simulator.
I can think of so many possibilities for it in Mastercam! Hopefully we see an integration soon.
To use automatic regions in Mastercam? :o
Why not build a language model of all the cnc programs you have run on a given machine and the manual for the machine. Can the bot understand 3d models?
One of the first things I did with chat got was ask it for some code, it was basically correct but would not have ran on a machine. I of course praised its response and that it should not change a thing. No need to train my replacement. 😉
Can it find the mistakes in its own output once told it has made a mistake?
Usually, yes.
I love the future. Now I'm waiting for a open source programm like airbus/apworks used to calculate that organic skeleton form of the light rider motorcycle.
PS: the earlier version of chat gpt was already pretty good with html-code.
I didn't try HTML but did have it do some VBA in Excel.
I use Co pilot to generated bas reliefs which I then process into STL files and carve out with the CNC
What happens if you put GPT generated code into GPT and ask GPT to improve it?
It will try and improve it. If it doesn't know what to improve, you may get odd results.
Would love to see you create a GPT specifically for gcode and similar applications!
How are you accessing v4 when 3.5 is the only version available for free online?
You have to pay for the extras. I think its $20/m.
how do you get the add image part on the text input section? I have gpt 4
me too and i can`t
I think its only available for GPT Plus users right now. Not sure when it rolls out to general GPT. Basically, you need to pay for it, $20/m.
It has a long ways to go! And also needs better instructions from operator....T1 = spotdrill (not T3-as at 3:09......this can confuse the logic of the GPT, unless specifically instructed to not follow tool numbers in ascending order-which is common practice. For chatGPT to understand that, it would add another logical hurdle and to overcome it would require additional instructions)....Maybe one day in the future...🙂
Where is the terminator robot on the thumbnail ?
I thought decent mostly automated 3axis milling has existed for a while. Isn't that literally protolabs whole business model?
I think their automation is more the quoting end, though I'm sure they have in house automations setup...none of which you would have access to...and thats where ai like this helps everyone.
@@CamInstructor I've been told by people in the industry that their 3axis cam and 2 axis turning is 90% automated, generally just reviewed by humans or humans program specific requirements. But it wasn't from prototolabs employees, might be total BS.
Gotta love the tech
I applaud you for exploring new ways of accomplishing tasks. The problem with AI is there is no I (intelligence). There is no creativity. No digital computer that by definition can only make decisions with on/off will never have intelligence. The intelligence is provided by the programmer. AI is nothing more than a super fast search engine, that goes through a huge database, and then outputs information based on rules defined by..... a programmer.
The computer that was trained to play GO beat the best players in the world using strategies no one had used before. It taught itself how to play, programmers did not. Same goes for chess, computers are better than humans. AI might be behind humans in many aspects, but the gap is closing quickly.
You should have uploaded a screenshoot of the incorrect result back to chatgpt to see if it could correct its error based on the image.
Alternatively you could have told ChatGPT that the bolt hole pattern looks like a square or diamond shape and to try again.
It will probably be a little while longer before chatgpt can one-shot things like this.
I did some back and forth trying to get it to correct the hole locations, even tried as you suggested, uploading an image.The image I showed it had the red circles in the correct location and black points were its locations, I then asked it if it looked correct. It was able to see it was not correct and attempted to fix. Still had a few holes off but interesting none the less.
@@CamInstructor At least ChatGPT tried, even though it still could not get it right. I am a chatgpt plus subscriber, but still have not gotten access the new image input feature. I was probably one of the earliest subscribers to chatgpt plus, but it always seems like I am one of the last ones to get access. Very frustrating. I have serveral things I would like to try as well.
very interesting mike!
Many thanks!
Music used?
Neon Renegade - Harris Heller
Honestly it just needs a feedback loop- for instance just sending the generated image of the generated locations vs the expected and see if it can spot it owns errors
It will get better, and it will get better quickly.
ChatGPT is my coding partner when I do C# or C. I do not use it for G-code. Same as normal code, it still goes off in strange tangents sometimes and forget some options available and you have to prompt debug it a lot so I feel for G-code the convectional tools are still more headache free than AI. In a generation or two down the road, when it have millions of tokens is will preform much better on all fronts. And it is really bad at making STL files too. lol. It will try but it will fail. haha.
We are well on our way to a world where not only can AI make the design for parts and whole machines, but actually design them from the ground up and improve on (evolve) the design over time. This means that there could come a day when machines entirely designed by AI are so advanced - after potentially thousands of iterations - that us humans would not even be able to understand how they work and would not be able to repair them when they fail.
Tip. If chat gpt makes a mistake and you know what the mistake is, tell it to fix its mistake explaining where it went wrong, ie order of operations and it will correct itself.
Interesting would be if you can do the same but feed him with pure text from DXF format, or even better with some text formated 3D model 😁
You can already do that with text input to describe a 3d model, that can be programmatically created using openscad(text based cad)🤔
Using AI can you produce a G code programme from a solid model ?
These language models, no. There isn't a way to give them a solid model....yet.
Do the creators of chatgpt just put its self to work on its self to fix its self?
It can’t use GibbsCAM
Most of my experience has been with 3D printers and CNC engravers, but I'm pretty sure most CAD software has gcode outputs...
Edit:
Okay, so this is a bit different, but if you wanted to be able to auto convert drawings into g-code that shouldn't be a particularly challenging thing to accomplish
I think it is worth mentioning that you have to turn on the the feature to allow uploads, currently in Beta, otherwise this video is confusing to a new user.
I didn't need to turn anything on to allow the use of images. Beta features only control plugins, bing and advanced data analysis.
@@CamInstructor I'm not sure then, I had to enable this and I have the 'plus' subscription service, otherwise I don't have this functionality. I'm using Google as the default browser, I don't know if that has anything to do with it. I had chatgpt4 for quite a while before signing up for the plus version so that maybe also be a factor also. Not sure what to say.
It didnt have the correct values with the 36sin, Black was the chatgpt numbers.
It had the correct formulas but somehow screwed the calculation on some.
@@CamInstructor I was mistaken, you are correct, I'm sorry I thought you were thinking it had the right coordinates.
By the time you give chat gpt all the details you could have Cam make the program. Plus you didn't specify center drill, end mill size and flutes
Wasn't really meant to be a head to head test for whats faster. More just to see what it can do. If its able to do it, it will eventually be the faster method, just a matter of when.
I have an interest in GCode, GpT4 and manufacturing. My major issue is I don't have a brain. I am not an engineer. I want to work with an engineer(s), so ideas can be produced from ideas I have. I have been waiting for the news you have brought, but I am intriqued with type of errors of precision being made by the high level Machine Language program.
As you have demostrated most problems can be debugged, and straightened out for correctness. This is obviously what must be done. More fascinating is the LLMs that are open source.
As you know, NASA has decided they are going into real estate on the Moon. Nasa has an itch to put a residential area in outer space. That is fine, but they are going to need some help.
And based on your video, you have shown vital information that will allow machines to be built on the Moon and not on Earth. Basically, as one scale's up, it would be a good thing not to write home saying we need this and that, when such a thing can be built by material that exists on the lunar surface. Very good job on the video production . Well done.
try making electrical schematics for breadboards
my only try using AI and G-code is this: I ask chatgpt to draw heart. it draw two semicircle over V. Clever solution.
I do not use AI in my daily engineering work (matlab/simulink in my case). But I feel soon I will, coz they are really going to advance it in a year or so.
Most CAD/CAM packages on the market can already automatically generate tool paths. FWIW, I wrote programs during the late 80's that ran on Z-80 CP/M systems that could read DXF files and apply MY logic to automatically write programs for Bridgeport milling machines and Strippit-Houdaille turret punch CNC's. The advantage of my code was that it always worked vs. the garbage that modern AI generates.
your work was and is probably very good, but you have absolutely no understanding of what this tech is capable of based on your comment.
@@LyonScapeStudios AI is capable of producing very low quality code, whether that is for CNC machining or writing software applications.
I've been testing AI's ability to write software applications. While the output would seem amazing to someone who had just graduated college with an advanced degree in computer science, it is readily apparent to a skilled professional that even the best of code generated by AI is riddled with defects, including design patterns which will limit performance, stability, and scalability.
I see AI for what it is - a con game that can be used to deceive the gullible.
@@gaiustacitus4242 thats your problem. You are assuming it will write code because thats what you do. It will not, it will just utilize it. Like an ape, it will use the sticks and stones around it to complete tasks. And preliminary testing has been stellar.
@@LyonScapeStudios The entire purpose of the video was to demonstrate the capability of AI to automatically program CNC machine tools. Likewise, I've watched other videos and read articles touting the ability of AI to write software applications for computers and mobile devices.
I'm not "assuming it will write code." I'm testing the claims, and so far I've only been able to dispel them.
AI is not supposed to be taking existing "sticks and stones" and use them for assembly of its output. The claims for AI tout its ability to LEARN how to generate NOVEL works.
As demonstrated in this video, AI's efforts to generate an equally spaced hole pattern used the correct formulae but it failed to properly apply it when calculating the X/Y coordinates.
AI also has a problem with using conventional milling for CNC cutter paths, when even a neophyte knows that climb milling is required for CNC operations (even if ignorant of the reason for doing so).
You look at AI and see it through rose colored glasses. I look at AI and laugh because while it is Artificial it most certainly has no Intelligence.
@@gaiustacitus4242 chat gpt is the tip of this iceberg. If you bothered to do any deep diving you would know that. There are many manufacturers worldwide that have completely automated programming solutions and they are very good. Please do some more reading before you take such a strong stance.
Slightly surprised that ChatGPT doesn't 'get' trig: Sine and cosine range between [-1,1], the expressions ...,1.5 + sin(36 deg) should have read 1.5 x sin(angle), multiplying the radius by the value of the circular functions, not adding them.
That's not what it did wrong. You can't see it in this video but the radius of the bolt circle is 1 so it's a unit circle with a center point of (1.5,1.5). The expression 1.5 + sin(36 deg)
is correct and that scares me even more. It should have been correct.
@@danarrington2224 I asked it to explain the formula it used for that location, hoping to see where it went wrong. While it was explaining, it realized it gave an incorrect value and then apologized lol. Asked it to double check all the locations and let me know how many it had incorrect...it found a few more but even after some more back and forth, it continued with at least 2 incorrect locations. Not sure why.
@@CamInstructor That's wild. If you ever do figure out why it did the math wrong, please let us know.
@@danarrington2224 Ah - I mistook a "3" I saw earlier (found the frame at 0:51, for the square drill pattern) for the later example's diameter Bolt circle of 2" dia on 3" stock (at 4:07). Got it. That IS odd.
i think the issue here is... people expect chatgpt to instantly know what u want based on random words u input which is ridiculous cause honestly talking to a human if u tell him concise words and sentences to the point most humans dont understand... so expect less from a machine that learns and does 100% exactly what u ask it... so if u ask something wrong... it will give u something wrong... i got the best results from chatgpt by explaining in detail everything i wanted and some possible issues also... so it can think of them and take them in calculations... and 99% of the time worked fine... there is that 1% when as u said... mid talking with it... forgets stuff... or just disregards it... cause if u ask it... based on the input above or first thing i asked.. it knows... so its all user input fault honestly... its a matter of understanding how to perfectly work with it... i find best results is give it small fractions of a big thing u working on... dont give it a full bulk load... lets say u can split a cnc program in 10 steps... ask it to code that entire project but ask it in 10 small parts drill some holes, do some chamfers etc etc... and at the end u cut off ur programming time by more then 50%. u can also tell it to skipp math lessons, just to give u the results as u are not interested in a explanation.
Great... now we're teaching skynet exactly what everything looks like not just a description
Excellent
Wait, it got the order of operations wrong....... On purpose maybe?
It is learning to replicate a body for itself! It's inevitable!
you need to ask it to use radians
It had the correct formulas and values. Not sure why it got the answers right for some but wrong for others.
Holy shit is all I can say. The speed that AI is evolving is a bit unnerving.
It takes too long to specify all parameters. You have to check the result in a CAD and correct it with AI perhaps multiple times until you get the satisfying result. AI is good in learning but it cannot learn or predict what the final product will look like which exist in your head only..
I thought this already exists, I have used this feature in mastercam
If you upload a screenshot of the red and black circles and then briefly explain it’s error, it will recognize it’s mistake and fix the code.
I tried this. It fixed some holes but still never got them all correct.
Must be the trig throwing it off. I was doing some TDAmeritrade Thinkorswim/Thinkscript programming. The program gave errors at first, but once I added screenshots, it fixed them right away. Could likely be fixed by an app call to WolphramAlpa.
This is crazy. I had asked chatgpt to write some code for a haas mill, and it said it did not have a model for the code. That was a year ago. I even asked Ai to add the code for a robot to unload and load a new part.
Hold on tight the skill gap is closing
Nice!
Thanks!
Could anyone here do sin of 36degrees in their head without a calculator? If you allow the Language Model (not math model) to use a calculator, like you did, it will get the calculation correct.
It got some correct though.
Indeed, the model is able to use knowledge of prior calculations of the sin of 39 degrees, the issue is that ChatGPT uses a non zero "temperature", this value can vary from around 0.5 to 1 with 0 being almost deterministic recall and 2 being whacko off the wall choices. With that in mind you can see how the models recall when exposed to this randomness (artificial spontaneity, lets call it) can go a little off. I'm a community champion over at community.openai.com and you're super welcome to drop by and I can walk you though maybe creating some API code where you can specify temperatures and other fine tuning of parameters, and even get you started with the Playground :D Always exciting to see new users and CNC was very interesting to watch.
1) Use the "Advanced Data Analysis" it is better for coding and can actually write and run Python code including drawing and simulation
2) Ask it to do "step-by-step" it backtracks and reread what it has done and often time it improves to an extent;
3) Invest in AI companies funds, in a few years there will be nothing that you can do on a computer that AI will not do better, faster and cheaper, it will only get smarter from here and, at a certain point there will be no way to make money other than by investing and some niche jobs. The same laws of physics applies to us and machines and given enough development, the machines will eventually take over.
Will be interesting to see where this goes in the near future
I don't know about you but we were able to do this in y2000 without AI when I took my edu
Image recognition is brand new. Yes, lots of machines were/are able to work with dxfs to create simple 2D paths, this is much different and a glimpse of where things are going.
Interesting but the way it operates so far is not conducive to such exact logic work. You might think that it fails on simple tasks but it's not using a calculator like you are, it's just doing the calculations on 'intuition' of math. Try answering your own prompt as if you were chatGPT and write it sequentially without ever going back and editing and reordering and without using a calculator. No notes on paper, just write it out immediately in your head.
So it's not quite ready for being used in the real world but it's glimmers of amazing capability.
Oh this is just great...now they can start designing their T-800 Bodies. Thanks for the end of the world !!!
But other than the beginning of the destructiion of Mankind, this was a Great video !
That thumbnail looks like a Megadeth album cover
DALL-E 3 created the image. Definitely has a Megadeth look!
Don't you need more information, HSS or Carbide drill bit, # flutes & type of endmill , coolant? Good start but your programmer or machinist is left with fixing the program, so I rather have a skilled person do it in a CAM software package, and who is doing the machine set-up? This might help students get a basic understand of G-code and using AI but when I teach I have to get students to understand G-code (adjust program code if needed on shop floor) and the different software used in machining plus all the tooling, fixtures, and the touch offs, etc. Like the video its a good lesson on AI advancing in manufacturing processes such as CNC machining.
Using AI generated gcode is asking for trouble
Today, yes.
@@CamInstructorForever, yes.
But do you get more or less trouble than with human coding? That's the break point.
cool
They fear AI CAM there will be no way to control humans 😂
Ai is just complex matrix math to design algorithms to produce data from a dataset. What jobs ai creates or takes away is a natural part of human development. For all of recorded human history, horses and blacksmiths were an essential part of human society that was completely replaced in two decades. But now we have more mechanics and engineers and business people and all sorts of jobs for people because of the car industry. There are more people working in the jobs cars created than cars ever replaced. Same with computers. There are more than double if not triple the human population on this planet than when horses and blacksmiths were phased out. Ai will make more complex and fulfilling jobs more accessible to those who might have been limited to less fulfilling jobs in more limited capacities. AI will not replace many jobs, but people using ai will. Learn these things now or you will be replaced. Any industry and job will require you to stay current to stay in that job or industry. AI is just the catalyst this time.
100%
so cnc industry is getting a hit danm
I think CNC will be one of the last to go.
the are putting nurolinks soon on people, i give it a good 5-10yrs . find alternative works soon. :( @@thesoloveichiks159
its quicker just to write a program on fusion
I gave chat gpt to cnc a BMW but it came up with an AUDI
0:31 you have to upload drawing if want ai programmed do it no intelligence. now your top secret drawing is with them. sure they want you upload like apple want your files on cloud
As long as the Ai is on servers that are ITAR compliant, you're good right? lol
I have not used chat gpt in like at least 6 months.
The fail with Gcode and math itself is well, it might seem underwhelming. But consider this - people who wrote the code for ChatGPTs AI - most probably have NEVER HEARD OF GCode!!!! Let alone understand it. Probably same is true with trig stuff.
I try this to shorten my program
where in the world is carmen sandiego
beside you
Not good enough. I don't want to make a drawing in the first place. I want to take an old part from my car, put it in a 3d scanner that is integrated into the CNC cabinet and have the CNC machine spit out a new part BASED on this that is in accordance with the specs and NOT a direct copy of the used part. So basically it must identify the part, look up the specs online, and just make the part WITHOUT me worrying about looking over GCODE or other stuff.
So you are applauding ChatGTP's ability to translate a completed design into CNC instructions. In Computer Science, we call this translation. Fortunately, the process requires no intelligence or creativity. It has been studied since the 1950s.
The creativity and intelligence are required to generate the design in the first place. (I will admit that most "designs" are minor modifications of previous designs. Creativity is very very very rare. Too bad there are no Expert Systems to help with this process.)
You have this completely backwards. AI is already able to do design, look at AutoDesks Generative Design. The machining is the hard part that it cannot do, yet.