In only 3 layers we get there. I think with more layers and new information searching + many more ideas I have, we can make mechanical engineers 1000x more effective across the board but skipping from need to idea for people who work with wide varieties of products etc. this is an infant version of what can likely be expected with the next gen versions of both GPT, Midjourney, and of course maybe even some Dall-E 2 for great style masks.
3 Likes
I think the key here is using LLM information pools and reference pools to expand on a simple topics and allow it to pack together highly contextualized information to generate an image from.
Another key idea here is that there no user involvement beyond just asking for something in simple but clear terms
really interesting approach of how to use this technologies for design, did you tried to ask about the command necessaries for a solidworks design of your mechanism?
1 Like
I think the amount of “thought” that goes into commands is actually VERY small, especially because SW allows for scripting that could autotemplate common parametric designs that it could then tweak and optimize for the application. I’ve got 12 years work experience in ME/EE/SWE and a PE in thermal fluids if you want to talk more about all this.
I think simple designs could around 100k tokens or something, and depends a lot on the design and the constraints which could all just be assumed at random and see what it comes up…
1 Like
i tried to merge your prompts with other prompt that i usually use for coding, maybe it is possible to define each part of the prompt to call to specific tools for test and create new products, maybe with the plugins feature creating personal calls to plugins in the input/output functions os a prompt
1 Like
