I've been asking chatGPT all sorts of questions about building science and the responses have been unbelievable. It appears to have the understanding of a 30 year industry veteran.
@@HomePerformance @RustyNail600 I view it as a kind of autocomplete over everything published ... so if it responds with a 30-year industry veteran knowledge, it also shows which knowledge is not present in this veteran: 1) zoning is complicating mechanics and causes more issues than it solves, and 2) even veterans don't understand pressure relief.
The most important part of chatGPT and other AI machine learning LLMs, is not a one off question like this, but rather to explain where it is wrong. Your next steps should not be to re-ask the question with "air tightness" for zoning in mind, rather, you teach it as a child who can learn instantly...with a response like "for point 7, zoning, the house will be very air tight, this will allow a more uniform climate across rooms. Would you change anything with your initial recommendation keeping this in mind?". A good power of this tool is it's ability to learn throughout a conversation, and remember this information. Don't start over like you would with something like midjourney.
Yes, that's really good. Sometimes you need to do a rewrite though just to stay precise within your token limit. As you add to the conversation, it reaches a point where it "forgets" about the details of the beginning and uses a summary of what it's forgetting rather than the exact words so some context and detail gets lost. There's a place for both adding clarification later and rewriting. Learning to be good at prompt engineering is critical to using this tool.
The biggest problem in getting a really good HVAC system isn't the general design it is the specifics of how to do it and then finding someone who is skilled, DILIGENT, and honest in not cutting corners to installed
It's amazing to see how AI, like ChatGPT, can handle complex math problems with numerous unit conversions. For instance, it can process manual J data and efficiently size dehumidifiers to offset latent heat load. However, as with any automated system, it's essential to be familiar with the manual calculations to ensure accuracy and to catch any potential errors or hallucinations made by the chatbot. Fascinating technology, though!
You mentioned that zoning isn't necessary in a tight house. How so? It seems counterintuitive to heat/cool the entire house when you are only occupying parts of it.
Because the solid stuff in the home reaches a stable temp, and it's actually a pain to try to let it drift away, plus you have danger of condensation if it gets too far off-target. Watch this: ruclips.net/video/Y0LXsGuSQFE/видео.html
Great video. I was trying to find the link to the video you watched twice about large language model AI but I don't see it in the description. I see the ChatGPT suggestions in the description though
I think I interpreted the filtration/purification part of the response differently than you did. How it appeared to me is that it was recommending a replaceable HEPA filter for the AHU, but was also suggesting a washable pre-filter to catch the pet hair and prevent it from clogging up the expensive HEPA filter. Then it was additionally recommending portable air purifiers to be placed in the rooms that those with health problems use most frequently. That part might not be necessary with a super airtight house, but like you said, it may not have understood that airtightness was part of the design goals. If you look at the results this way, I would say that the AI did a pretty remarkable job with the exception of the pressure relief bit. I heard they were incorporating a plug-in for Wolfram Alpha into the latest version of ChatGPT, so if given the right information it might even be able to do the necessary calculations to determine HVAC, ERV, AHU, and ductwork sizing in the near future.
Yep. Prompt engineering is critical. I've been working with ChatGPT and have found that knowing how to structure your query is absolutely essential to success.
This does seem like pretty standard hvac information that it's just generically regurgitating. I'd like to see it size some ductwork for all the said components and put together a coherent layout. Of course there is software that assists with this already (Wrightsoft, etc), but is quite clunky and fairly manual to use (at least since I last used it 6 years ago). I don't expect AI to replace your job, but I do expect AI to speed up the workflow of designing complex systems: think more of a drag-and-drop, stretch this trunk, delete that run type of interface while it auto-updates everything up and down the line accordingly. Duct sizing/load calcs are fundamentally math/data-based - which is perfect for computing. If you know your desired input, know your desired output, and know distance and friction losses in between - everything else can be calculated in real time. That would be ideal. The experienced hvac professional is the guide and the software does the grunt work nearly instantly.
OpenAI is the official chat GPT. All others are using a similar language model but it’s not the official. Doing this in OpenAI would give you a more intelligent answer and would be much faster.
Hi Corbett, using artificial intelligence to design an HVAC system is an interesting concept. If I were you, after seeing the suggestions made to your query I would have immediately re-run the design simulation and included the information about the air-tightness of the building envelope and the space zoning requirements (and perhaps data on the climate zone) and see what changes the AI program would make to its previous suggestions. Can this program also make specific equipment recommendations or design the actual layout for the duct system and mechanical room?
Thanks for sharing. ChatGPT is truely astonishing yet also difficult to seperate fact from fiction as it interpolates the internet. Getting better very fast. The diagram you showed has a number of fascinating suggestions. I particularly noted dedicated returns per level, ERV integrated into main duct system for supply, and dedicated humidifier supply. The former is new I think, while the latter two you've touched on but here you seem to be explicitly recommending. Would love to hear more on your thoughts.
Yes, I have a boxing mouth guard in, I'm about to go ten rounds with your mom. It's invisalign making my mouth high performance. Forgot to take them out.
AI is fascinating and scary. How many jobs it will eliminate is something of concern for knowledge workers that could be replaced by it to some extent.
Our worry is right now we have a child, that isn't able to destroy the planet when someone asks, 10 more years it'll be asking some someone to solve the captcha that secures the missile silos.
At the end of the day, AI is still just a glorified calculator like the original computers, a series of binary 0's and 1's that spit back at you what it has been preprogrammed to do. That should always be our starting point when we begin to worry about technology. Not to get into the whole 'problem of consciousness' debate, but I think it's wrong to view this thing as a child. I think it would be more appropriate to view it as a large, dangerous piece of heavy machinery that would chop someone up and have no 'realization' of what it had just done, no more than a wood chipper could. Like heavy machinery we should put all the redundant safety guards on it we can, train people what they're working on, and hope for the best haha.
![I Replaced ALL my ADOBE APPS with these [free or cheaper] Alternatives!](/img/1.gif)
I've been asking chatGPT all sorts of questions about building science and the responses have been unbelievable. It appears to have the understanding of a 30 year industry veteran.
Everyone just got smarter (hopefully)
@@HomePerformance hopefully. The concerns about AI are real though.
@@HomePerformance @RustyNail600 I view it as a kind of autocomplete over everything published ... so if it responds with a 30-year industry veteran knowledge, it also shows which knowledge is not present in this veteran:
1) zoning is complicating mechanics and causes more issues than it solves, and
2) even veterans don't understand pressure relief.
Exactly- we need to put more nuanced info out there so it learns to incorporate the latest thinking. A lot of outdated opinions to outweigh.
The most important part of chatGPT and other AI machine learning LLMs, is not a one off question like this, but rather to explain where it is wrong. Your next steps should not be to re-ask the question with "air tightness" for zoning in mind, rather, you teach it as a child who can learn instantly...with a response like "for point 7, zoning, the house will be very air tight, this will allow a more uniform climate across rooms. Would you change anything with your initial recommendation keeping this in mind?". A good power of this tool is it's ability to learn throughout a conversation, and remember this information. Don't start over like you would with something like midjourney.
Interesting point, James
@@HomePerformance I don't mean to tell you what to do...I re-read and it sounds like I'm ordering you around.....
No way man, I hear you loud and clear :)
Yes, that's really good. Sometimes you need to do a rewrite though just to stay precise within your token limit. As you add to the conversation, it reaches a point where it "forgets" about the details of the beginning and uses a summary of what it's forgetting rather than the exact words so some context and detail gets lost.
There's a place for both adding clarification later and rewriting. Learning to be good at prompt engineering is critical to using this tool.
Great video! The specification statement alone at the beginning was worth the watch. The ai results were surprisingly good.
Thanks Jeremy
The biggest problem in getting a really good HVAC system isn't the general design it is the specifics of how to do it and then finding someone who is skilled, DILIGENT, and honest in not cutting corners to installed
It's amazing to see how AI, like ChatGPT, can handle complex math problems with numerous unit conversions. For instance, it can process manual J data and efficiently size dehumidifiers to offset latent heat load. However, as with any automated system, it's essential to be familiar with the manual calculations to ensure accuracy and to catch any potential errors or hallucinations made by the chatbot. Fascinating technology, though!
You mentioned that zoning isn't necessary in a tight house. How so? It seems counterintuitive to heat/cool the entire house when you are only occupying parts of it.
Because the solid stuff in the home reaches a stable temp, and it's actually a pain to try to let it drift away, plus you have danger of condensation if it gets too far off-target. Watch this: ruclips.net/video/Y0LXsGuSQFE/видео.html
Great video. I was trying to find the link to the video you watched twice about large language model AI but I don't see it in the description. I see the ChatGPT suggestions in the description though
ruclips.net/video/xoVJKj8lcNQ/видео.html
Sorry about that, fixed.
I think I interpreted the filtration/purification part of the response differently than you did. How it appeared to me is that it was recommending a replaceable HEPA filter for the AHU, but was also suggesting a washable pre-filter to catch the pet hair and prevent it from clogging up the expensive HEPA filter. Then it was additionally recommending portable air purifiers to be placed in the rooms that those with health problems use most frequently. That part might not be necessary with a super airtight house, but like you said, it may not have understood that airtightness was part of the design goals. If you look at the results this way, I would say that the AI did a pretty remarkable job with the exception of the pressure relief bit. I heard they were incorporating a plug-in for Wolfram Alpha into the latest version of ChatGPT, so if given the right information it might even be able to do the necessary calculations to determine HVAC, ERV, AHU, and ductwork sizing in the near future.
Good point RJ
Follow-up questions and feedback can greatly increase output. As well as longer prompt. Prompt engineering is going to be a whole new field
I have found some of my most valuable insight giving it input and then asking it to critique it or why this is a bad idea.
Yep. Prompt engineering is critical. I've been working with ChatGPT and have found that knowing how to structure your query is absolutely essential to success.
This does seem like pretty standard hvac information that it's just generically regurgitating. I'd like to see it size some ductwork for all the said components and put together a coherent layout. Of course there is software that assists with this already (Wrightsoft, etc), but is quite clunky and fairly manual to use (at least since I last used it 6 years ago). I don't expect AI to replace your job, but I do expect AI to speed up the workflow of designing complex systems: think more of a drag-and-drop, stretch this trunk, delete that run type of interface while it auto-updates everything up and down the line accordingly.
Duct sizing/load calcs are fundamentally math/data-based - which is perfect for computing. If you know your desired input, know your desired output, and know distance and friction losses in between - everything else can be calculated in real time. That would be ideal. The experienced hvac professional is the guide and the software does the grunt work nearly instantly.
What are your thoughts on the sprays for fabric to eliminate odors in your home?
I heard it’s coming out with its own home Performance show on pbs.
Haha oh my god please, this is making my hair grayer every year
OpenAI is the official chat GPT. All others are using a similar language model but it’s not the official. Doing this in OpenAI would give you a more intelligent answer and would be much faster.
Very cool video! Thanks
Great video!
Thanks Hamp
Hi Corbett, using artificial intelligence to design an HVAC system is an interesting concept. If I were you, after seeing the suggestions made to your query I would have immediately re-run the design simulation and included the information about the air-tightness of the building envelope and the space zoning requirements (and perhaps data on the climate zone) and see what changes the AI program would make to its previous suggestions.
Can this program also make specific equipment recommendations or design the actual layout for the duct system and mechanical room?
I think you just gave yourself a homework assignment, Ryan
Lets give it a layout of a few rooms...see what it comes up with..
Thanks for sharing. ChatGPT is truely astonishing yet also difficult to seperate fact from fiction as it interpolates the internet. Getting better very fast.
The diagram you showed has a number of fascinating suggestions. I particularly noted dedicated returns per level, ERV integrated into main duct system for supply, and dedicated humidifier supply. The former is new I think, while the latter two you've touched on but here you seem to be explicitly recommending. Would love to hear more on your thoughts.
Thanks for watching, Vahid. Stay tuned, I pretty much talk about those topics nonstop.
You using a mouth guard or Invisalign by chance, or just my ears hearing something a little different
Yes, I have a boxing mouth guard in, I'm about to go ten rounds with your mom.
It's invisalign making my mouth high performance. Forgot to take them out.
AI is fascinating and scary. How many jobs it will eliminate is something of concern for knowledge workers that could be replaced by it to some extent.
Yes, it looks like liberal arts college is out, trade school is back in. As far as professional training goes anyway.
Why is zoning not necessary with a tight house?
HVAC Zoning: Do You Need It, or Not?
ruclips.net/video/Y0LXsGuSQFE/видео.html
Our worry is right now we have a child, that isn't able to destroy the planet when someone asks, 10 more years it'll be asking some someone to solve the captcha that secures the missile silos.
Glad I can still sling ducts…
You’re not fired then. You’re rockstar level anyway, nothing to worry about.
At the end of the day, AI is still just a glorified calculator like the original computers, a series of binary 0's and 1's that spit back at you what it has been preprogrammed to do. That should always be our starting point when we begin to worry about technology. Not to get into the whole 'problem of consciousness' debate, but I think it's wrong to view this thing as a child. I think it would be more appropriate to view it as a large, dangerous piece of heavy machinery that would chop someone up and have no 'realization' of what it had just done, no more than a wood chipper could. Like heavy machinery we should put all the redundant safety guards on it we can, train people what they're working on, and hope for the best haha.
In the near future architects and engineers to be replaced by chatGPT