sent ya a PM 
wish i had the room 
3 Likes
8 Likes
I donāt see this leading anywhere, specifically because the input medium is so very vague. Not as a function of the tech, but simply the limits of stringing together single words into chains big enough to mean anything.
Currently we use CAD software to design, which gives us a 3d image. Pre-computer, I could have hand drawn an image which would contain a great deal of information. The medium contains massive detail in a small package regardless of tech.
Now theyāre trying to use the written language to replace it. Ever try describing something in complete detail? You can hit the major points, but youāll never be able to get all the major ones much less the finer specific points.
This is why we have wanted posters instead of wanted descriptions.
For example, compare the two.
- White, Male, mid 40ās, short hair, glasses, with short beard and sandy colored curly hair, wearing a t-shirt.
Try looking at what google images does with that prompt.
Any description accurate enough to convey exact meaning would make the Encylopedia Brittanica look like a footnote.
4 Likes
Then try limiting it to 60 characters. Thatās my gripe with the software
3 Likes
If it were something like āMake a desk statue of super mario holding a ā1 upā mushroomā it might be okayā¦ But āmake a case for a raspberry pi 4b with 4 3mm screw holes in place for a 30mm fan centered above the cpu with 10cm clearence and a slot above GPIO pins 1-8āā¦ Yeah no
And this even assumes it knows all of the exact size specs for an rpi 4b
2 Likes
I meanā¦ this is exactly what people were saying about basically all AI just a couple years ago. Thatās why I recorded the video. Iām sure the quality of output will be so much better fairly soon. Maybe one day you could just say āmake a case for a raspberry pi 4b with 4 3mm screw holes in place for a 30mm fan centered above the cpu with 10cm clearence and a slot above GPIO pins 1-8ā
Thatās kind of my point though. If you had video and pictures of me for instance, but an AI hadnāt been trained on them, could you get it to produce an image that would pass as being me? Even if it had 16 fingers. You canāt use enough words to capture the level of detail, and still keep it human level language.
Yes I can see that this first edition of buildplate would not likely be able to create a perfect representation of you, but in a world of āclose enoughā thatās not really relevant for almost every purpose. I know this will grate hard against analytical / engineering brains but for the most part, AI solutions will be sufficient for the intended application.
Absolutely I could get it to produce an image that would āpassā as you. Even though the worst possible first generation AI representation of Will Smith eating spaghetti offered above is comically bad, the ideas āwill smithā and āeating spaghettiā are still conveyed to anyone who sees it. It succeeded in āpassingā the ideas being communicated. While this first edition buildplate AI thing is not likely able to do that even with good photos and videos of you, it wonāt be long at all. The AI tech needed to do it well already existsā¦ itās just a matter of hooking it up to buildplate or having these guys catch up their AI to fill in these gaps.
There seem to be two types of people I run into;
-
people that need to be given step by step instructions on how to do a task because they are unable or unwilling to reason out how to do a thing
-
people that have wide enough experience (typically driven by general curiosity) and most importantly the ability / willingness to apply that knowledge to derive / deduce how to perform a task (figure it out).
I think what people believe about AI not being trained on a specific thing means it canāt do it is relegating AI to type 1 people - they need precise exact instructionsā¦ but more and more we are seeing that AI is being layered to be more like type 2 peopleā¦ and eventually what we are grouping together as āAI reasoningā will be able to do an actual good job of filling in the gaps in its training / knowledge through web search and reasoning aloneā¦ and will have enough background knowledge and reasoning ability to actually do a good job completing tasks its givenā¦ just like us type 2 people do.
When it comes to complex tasks like making a case for a Raspberry Pi 4b with specific requirements, layers of AI tasks will definitely be able to handle this even if not trained specifically on those things. The thing many people donāt seem to be grasping is that analytical problem solving is already a thing for AI and will only improve. Getting the pi 4b dimensions and maybe even existing case STL designs out there and making the modifications requested is not that difficult if you just add a little bit of reasoning. Knowing how to build the case from scratch is not entirely necessaryā¦ but even this will eventually be possible by breaking down the steps to solving the problem into discrete subtasks and applying the entirety of its knowledge and reasoning to each individual subtask.
This is coming guysā¦ mark my words.
1 Like
haha ok this is comically bad, as expectedā¦ but surely anyone can see the potential here. I broke down the steps and simply asked buildplate āraspberry pi 4b caseā and it came up with this;
It should be clear now why this is the currently limitā¦ adding too many customizations through a longer prompt, at this stage, would likely result in absolutely useless nonsenseā¦ but with a simple prompt it got a vague idea of what was neededā¦ yes it included something resembling a pi in the actual STLā¦ but this is like baby stagesā¦ itāll grow up, and useful STLs will probably start popping out of this thing pretty soon.
They allow you to publish any of your results, and you can explore what other people have produced. Of course this is a curated list of what people felt was worth while publishing, but thatās my point ā¦ this is already producing useful STLsā¦ at least useful for the people that generated themā¦ most likely just art piecesā¦ a couple helmet designs look pretty descent.
Will I be using buildplate for anything in the near future? Probably notā¦ but thatās not the pointā¦ itās comingā¦ thatās my only point.
4 Likes
This is either one of those things where weāre talking cross purposes or weāll just have to agree to disagree ( a fabulous idea in todayās climate ). Iāll say this, I absolutely agree with regards to tech development, in fact you may even be selling it short, tech begets tech at a logarithmic rate.
My entire issue is the bandwith limitations of language (human style, aka english / spanish / etc.)
But, I like to put my money where my mouth is, soā¦
If in 10 years (an eternity in tech development), you can write a 1000 word or less prompt that causes an AI to generate a passable (recognizable as me to someone who knows me, not just similarish) image of me, on a system that does not know what I look like (no access to previous images of me), then I will buy you a beer. Or download it, if the tech has gotten that far. 
2 Likes
I played with it a bit and it has great potential. But if it was iterative. Like you can say.
Make a blank rpi4 case.
Add a top to it
Make the top screw in at all four corners
Make a place to add a 40mm fan in the middle.
If it was able to be iterative, then it would be a game changer
2 Likes
Everything can be described in languageā¦ whether thatās English or mathematics, it could be done. English is often used in lab environments around the world because as a language it is a little more precise than say Spanish, but the raspberry pi case example is a good one because there are a lot of specifics.
You could definitely describe the length and the width and the height and the radius on the corners and the thickness of the walls and where exactly you want holes of what diameterā¦ all of that is totally possible, especially when you break down tasks.
In the spirit of trading places and other Hollywood trope movies, I will bet you the grand sum of $1 
3 Likes
Deal, we have witnesses, see you in 10!
edit, just realized I bet a beer against a dollar on a 10 year inflation curve, in a highly inflationary forecast economy.
6 Likes
If bottle caps become currency in ten years you may have the better deal
3 Likes
No noā¦ clients wonāt be replacing programmersā¦ programmers will replace the tedious aspects of programming wo they can focus on the broader scale architecting.
2 Likes
2 Likes