Just so, you know, the AI (ChatGPT, Gemini) is able to generate code for CCX/CGX. I was not able to get working code from the first request but at least it starts working after few corrections.
I assume, that this can be useful at least for education.
Also ChatGPT “told" me” that extended version is able to run and debug input files, for CCX installed on remote instance, such as Amazon EC2. It will generate the code, run it, read output, correct and run again until it will obtain solution.
I have a little experience with AI, just playing and trying to understand how it can be useful.
Can somebody share similar experience, regarding Calculx?
ChatGPT may handle basic input decks but it tends to make up Abaqus keywords in even a bit more advanced cases. It’s somewhat better when it comes to Python scripting (although makes mistakes and can’t really fix them in many cases). Thus, I would be very cautios regarding such tools and they are mostly useful for me to do some translations of large amounts of text, perform very basic calculations or just get a few ideas to explore further myself.
Basically, the biggest issue with AI is that it makes up things instead of admitting it doesn’t know the answer. Especially ChatGPT has that tendency. It even suggests non-existent books.
Based on our experience, we can confirm that AI tends to invent keywords and methods. Furthermore, we believe that CGX is a very powerful but underutilized tool.
Take a look at our news page dedicated to this topic:
Then, try asking an AI to create something similar. To create the mesh in the first image, we developed a special C application that, starting from general bellow dimension data, automatically generates the CGX input and the complete mesh.
I would suggest main Math, Physics and Engineering repositories and Publishers to start requesting that publications be fully understandable by IA. As a pre-requisite for the admision. IA should be able to respond about basic aspects of the content and should be able to read and reproduce correctly all the information including formulas, graphs and footnotes. Let’s help AI as if they were childs.
I am completely convinced that any scientific publication with the (AI Readable) stamp would be a resounding success.
By fully understandable i mean in a more basic level “fully readable”. Comprehension of the content is a second step that will be hopefully better and better.
Books well organized, structured, with readable formulas, graphs, symbols, units and definitions. Even stating clear the mathematical sign convention across the book.
In the same way you sometimes get uncomprehensible text when you try an OCR on a pdf, I can imagine how difficult can be for an IA to extract information from reference books and papers (even more the old ones).
For Example , I have tried to make IA to work based on well-known Roark’s formulas book. It’s almost impossible to get the IA to read the text and extract precisely any study case.
Not because it doesn’t know how to manage formulas. It is because the text is so chaotically organized that makes IA to mix variables and paragraphs from different areas of the book.
Regarding anthropomorphizing the IA. Yes, I approach it under that premise. We don’t want IA to dehumanize us in the future.
in the diagrams of paper mentioned clearly show knowledge based system (KBS), a crucial part containing math, physics and engineering experiences of developer. To my understanding, Artificial Intelligences is a human product not to competitive but helping to assist.
