AI Is Eating Software! What should we do?
ฝัง
- เผยแพร่เมื่อ 9 ก.พ. 2025
- They say that AI is eating software and that programmers will be out of jobs. Can AI design a house? In this video, I am giving my opinion on how AI changes our industry and what we should focus on.
Register at our gaasworld website and get access to a completely free, 10+hours Rhino Developer C# course:
gaasworld.com/
Follow our work and join our projects!
www.programmin...
LINKEDIN: / milosdimcic
INSTAGRAM: / milosdimcic
Thank you for your support!
Stay free!
#BIM #Revit #developer
Milos Dimcic
Dr.-Ing.
www.programming...
In my opinion, the main limiting factor here is that AI (specifically in this text I mean, LLMs ,image diffusion models and their combination) is trained on language and images, whereas a lot of the expertise of architects is locked away as "experience" that is not always well documented for an AI to train on. For this reason, I believe an AI may still struggle to write that middle layer of functions targeting the AEC industry and require us instead to work as developers to build simple functions to "Build a road", "Build a network from roads", "Analyse the connectivity scores of the network" - using which AI could certainly do everything you say. I don't think this limitation will stay in place for more than 5 years or so, but for now I think it is significantly less adept at "thinking" through architectural design challenges than it is at other fields that DO have ample training data (like medicine, art or law).
Curious to hear your thoughts! Ive always enjoyed your videos by the way!
Experience disappears because we have the result. If you think an LLM who can contextualize all available architectural data won't able to compete with your unquantifiable experience then i am afraid you're not ready to be a high agency person in an AI world.
@@GoodBaleadaMusic I dont think we have the "result". Its hard to find a large enough data set for architecture that contains interlinked buildings floor plans, thermal performance, privacy scores, construction methods and more performance metrics that can be reliably called "good" for a given context.
My only point is that it will take us more time to assemble such a training set of what "good architecture" is, whereas if you think about medicine or law "best practices" are documented pretty well in text. I do believe this is merely a delay in AI being good at architecture, not an insurmountable barrier.
@ashwin4991 I think you just answered your own question. I don't think we're going to trust humans with all of that. The central LLM will be the quality control. Everyone will get to design and probably more freely. Because all the architectural restraints can be built into All of your whims. You know? You can dream about a building in Europe from Milwaukee without having to contextualize too much of the local code. Because you can just throw your idea out and then an LLM will tell you how you need to augment it to match what your idea is for that regulatory framework in another country. I'm sorry but we're just not smart enough anymore. Not to contextualize everything that you just mentioned. You just made it sound dangerous to trust a human
Good to hear from you again! Really appreciate your insights.
Nice to have you around again.💥 I believe the true power of AI in architectural design will lie in its ability to seamlessly integrate sketch and hand drawing, effectively replacing and enhancing the UI level you mentioned. (Imagine software that intelligently interprets these intuitive inputs and translates them directly into BIM geometry). This to me, feels like a romantic notion: a return to hand drawing as the primary tool.
So we still stuck with revit?
Question: Can AI explain how the ancient Egyptians and Mayans were able to build their pyramids, given the technology that was available to them at that time? And what about AI explaining how the ancient Romans and Greeks were able to build their buildings, such as the Pantheon and Parthenon, again, given the technology at that time? It's simple enough to design a glass and steel box---which is what "architecture is today--but I'm quite impressed by what ancient civilizations were able to accomplish.
they can't. You're asking them a question they don't have an answer yet or solution to yet. Think fingerpainting, AI is doing finger painting so small, people see it as a great clear picture. Or think of AI as just spewing word salad til humans are "happy" with the solution. Cuz current AI doesn't know about room/house design without RL/or a AI model excusively for architech. Or AI doesn't know bodily physhics when doing AI video. Ai can't draw humans with correct number of fingers... without another HUGE work of RL to make them understand.
You want to do the design through successive approximation. For example, first off perhaps, 'show me ten popular designs of a 4 bedroom house'. Choose the one you prefer. Tell it to build according to UK building regs. Tell it you want a different type of brick and so on. You can just keep on making modifications large and small as you please.
Developing AI, is like raising baby Satan.
milos posts. i watch
2030: End user / customer to winner ASI: "Check our social media and design the house of our dreams, check with the bank if we can afford it and if so, get it built!"
AEC practitioner middleman: on UBI, fishing.
Your face looks skinny. I just watched you after a long time.