KyleBernhardt

@PermitZIPhvac @mwmoedinger @MrLGroves @TopherNOW @RE_Deal_Junkie I partly disagree. No question that a fine-tuned LLM deploys knowledge at scale unlike any piece of technology that came before it. Disagree that people just needs to do their homework. User experience of the tool needs to cater to the task first. People will then follow.

I see what you did there—clever, and I don’t disagree! Changing operational infrastructure to incorporate fine-tuning LLMs is foundational. Educating employees on how these models work—and making them work for you—is essential. Getting it right really does matter. AEC professionals want tech tailored to their needs, but they also need to understand it takes a similar commitment to studying LLM's needs to make them effective. It's not a one-way street, and so most revert back to archaic workflows. Most in the industry can’t differentiate between RAG, training, or prompt engineering, and many AI advocates don't even know what a "prompt injection attack" is—a serious security risk for firms. Some firms are putting in the work and paying attention. We're one of them. As an MEP consulting engineering firm (@PermitZIP), we understand nuance and know how to implement these tools. We build custom vector database pipelines, train our engineers to provide fine-tuning data to our software team, and ensure our employees know the limitations and capabilities of LLMs. We treat our in-house AI assistants like junior-level employees—they get better over time, but unlike people, they can read thousands of emails, texts, and files in seconds and generate useful insights. The issue isn't with tech bros—it’s that AEC professionals aren’t willing to put in the effort to truly learn these tools. The reason it’s not working is simple: we aren’t doing the homework.

GCs now taking 3 months to give *ballpark estimates* is killing me. I was an estimator - I KNOW it doesn't take this long. I can't work with guys who do this. During the bidding phase, we are sitting ducks - can't push the design ahead, so we can't bill. When estimates took 2-3 weeks, it was a blip in my cashflow and staffing. Now, they need 2-3 months to produce the same level of work, and they'll need that much time *again* when we're finished with the drawings?! These are intricate, intense custom renovations, in a tough market. These aren't run-n-gun slapdash remodels. But I've got guys who can turn estimates around in a coupla weeks, and guys who need 3 months...who do you think I'm going to put in front of my clients?? The guy who needs 3 months for ballpark and 3 months for final means he's stretched my project by 6 months....it's BS like this that makes these projects so expensive. Estimating is an art - and they're missing that, big time. Again, for the people in the back - I came up through the ranks, I DID that job - I know it's not easy, but it's also not a 2-3 month endeavor. Guys like that - no matter how good they are - will get left behind.

@PermitZIPhvac @mwmoedinger @MrLGroves @TopherNOW @RE_Deal_Junkie My dude you misunderstand me. LLMs are a probabilistic code architecture. I'm on board that train as a way to deploy expertise through technology. My critique is that LLMs powering chat bots are not the disruptor. It's the new technology experiences they make possible that are

You talk about probabilistic code architecture as if that doesn't describe humans. :) Architecture, engineering, and even building codes are all based on probability: Take the NEC tap rule—it's based on the likelihood of someone noticing a disconnect based on the distance from a tap. “It's probably safe enough.” Existing building codes also work on probability, allowing older systems to remain untouched based on the scope of work. We design HVAC systems assuming temperatures will stay within thresholds based on probabilistic weather data. We make technical decisions based on the probability they'll meet client needs—because it's impossible to consult on every specification. We're constantly making assumptions and predicting performance. The difference? We can’t process terabytes of data in seconds. Thinking this is just about a chatbot is precisely why the AEC industry will keep falling behind.

@PermitZIPhvac @mwmoedinger @MrLGroves @TopherNOW @RE_Deal_Junkie LMFTFY: "LLMs add scale to nuance when coupled with the right user experience that builds trust." LLM is one of the footings, not the house.

@mwmoedinger @MrLGroves @TopherNOW @RE_Deal_Junkie “Highly custom, idiosyncratic, specialized construction” doesn’t deliver on-demand, three-week turnaround estimates—unless you bring LLMs into the mix. LLMs add scale to nuance. Those who harness this with experience and solid processes will unlock superpowers.

@PermitZIPhvac @mwmoedinger @MrLGroves @TopherNOW @RE_Deal_Junkie I think you're a bit too focused on the technology approach here, and too little on the job and user experience a probabilistic code architecture could create. At the end of the day though, it's humans and businesses that will stand behind estimates, so a chatbot ain't the story

Time will tell! IMHO, those investing in LLM training now will soon be outpacing others by orders of magnitude. It's going to be tough (if possible) to keep up once that momentum builds. Let's hope the experienced GCs are the ones steering when that happens—for everyone's sake! Haha.

@Twiceroadsfool Goodnight, sweet prince.

What a day for us at TestFit and our partners at Prologis!

Remote work life hack: shamelessly talk to yourself out loud as you solve problems. Nobody's around to care and it helps organize things. Try it.

@Twiceroadsfool @katiecouric Come to the dark side. The sliding doors are worth it.

I don’t care about blue check marks. I mean, I drove a minivan for 15 years @elonmusk

@a2d1s @MiketMatheny @Twiceroadsfool @TravisUsesRevit @autodesk Fun fact: RL is part of SWC. First step of the process, actually.

@MiketMatheny @KyleBernhardt @Twiceroadsfool @TravisUsesRevit @autodesk even if they hit RL before syncing ?

Where does @Autodesk BIM360 Revit 2023 save the local copy of your model? Got a message that my ‘copy’ wasn’t current w/central when I came in this morning. Knew that I synced yesterday afternoon so ignored it. Evidently central didn’t have any of my work from yesterday PM.

@Twiceroadsfool @MiketMatheny @TravisUsesRevit @autodesk Totally agree with you here.

@Twiceroadsfool @MiketMatheny @TravisUsesRevit @autodesk There's a reason for that, although you may not agree with it -> the worksharing username needs to align with the Autodesk identity, so the change tracking in the construction cloud is always associated with a understandable name.

@MiketMatheny @Twiceroadsfool @TravisUsesRevit @autodesk Best solution is to avoid closing a session on a file with unsynced changes; regardless of where they are signed in. None of this is possible without that.

@Twiceroadsfool @KyleBernhardt @TravisUsesRevit @autodesk Could be. Conference rooms are an issue with some of our clients. Especially if they just use a generic login/password, but each user signs into Revit with their account. Not sure what the best solution would be there is? We saw it quite a bit at our past firm.

@MiketMatheny @Twiceroadsfool @TravisUsesRevit @autodesk That'll do it too.

@KyleBernhardt @Twiceroadsfool @TravisUsesRevit @autodesk Have seen users having a file open on their desktop, go to a conf room for a meeting and open the model there and sync in the conf room. If they go back to their desk and work on that open file I believe you get the same error message as there are now 2 local files for that user

@dourevit @Twiceroadsfool @AutodeskRevit @autodesk Totally agree. This is a meme-worthy error of an error.

@KyleBernhardt @Twiceroadsfool @AutodeskRevit No, besides the error state, it is very informative. But once something unexpected happens, all efforts to produce a good dialog become futile if not addressed :) Thanks for responding though, hope @autodesk can fix this. Cathches out users all the time.

@TravisUsesRevit @autodesk Don't know what to tell you man. It's literally impossible to get this error state from any other reason. Chalk it up as a learning experience. Next time you get the prompt, don't throw out your unsynced work.

@KyleBernhardt @autodesk In this instance then I would have had to do that ‘by accident’ multiple times.

@TravisUsesRevit @autodesk The save local button can be right next to the sync with central button, so it happens.

@KyleBernhardt @autodesk Hmmm. Good to know you can accidentally do that?

@dourevit @Twiceroadsfool @AutodeskRevit Yeah, I get that part. This is a just a bad error state. No way to sugar coat that. Thought you all had issues with the underlying dialog itself.

@KyleBernhardt @Twiceroadsfool @AutodeskRevit If you display 4 items and they all succeed, how can it be that it still failed? So either it's missing some other bar that should be red, or one or more of those four failed to complete and should not display 100% successful.

@brianmpayne @Twiceroadsfool @dourevit @AutodeskRevit Good feedback on that one. Retry would make sense in the failure case. There is already retry built into some of the steps (to handle connectivity issues), so why not do it for the whole command...