This episode reveals LandPricer.AI achieving 95%+ accuracy in image analysis using non-reasoning models for cost efficiency, with AI now correcting human labeling errors and identifying hidden structures behind tree cover that the founder initially missed—signaling AI will surpass top-tier land investors in aerial imagery analysis within 6-12 months.
Key Takeaways:
- AI corrects expert human image analysis errors Multiple test images showed AI identifying structural improvements and driveways that were mislabeled or missed entirely during manual review.
- Non-reasoning models achieve 95%+ accuracy at scale Using weaker, cheaper models (not reasoning-heavy ones) keeps per-property costs viable while maintaining commercial-grade accuracy across 30+ property characteristics.
- Six months until AI outperforms all land investors Current trajectory suggests AI aerial imagery analysis will exceed human expert capabilities by late 2025 across the entire industry.
Listen to the full episode for the philosophical implications of being surpassed by your own tools and how to maintain competitive advantage.
(Podcast transcript below)
Hi, Chris Duff over at Serious Land Capital, vacant land funding partner. Just wanted to give you an update on land pricer progress, specifically one aspect of it. I mean, we are really gunning to get this thing out within the next month here, June, 2025. I know it’s just been endless delays, but you know, product wasn’t where we needed it to be upon beta feedback earlier this year. Had to get back to the drawing board.
make it a lot faster, utilize new technologies that had come out and, you know, just realize, hey, we didn’t take the best approach off the bat. What are you going to do? Get back to the drawing board, figure it out, and, you know, utilize as much as you could from the first iteration and just improve it from there. So, you know, the UX was really what we’re trying to focus on. And so,
What I’ve been primarily building in the past few weeks, know, it’s foundational to the product is just removing a lot of the required user input that the previous version of LendPrice are needed. And just to shift that over to AI LLM models, large language model for the acronym there, in case you weren’t familiar to handle a lot of
the IDing of various features and steady progress has been made there. Like I really want to get it to the point where the prompts, the image prompts are identifying or getting the correct answer over 95 % of the time. Most of the questions.
like the 30 plus that we’re working with at the moment are already hitting that threshold or very close to it. and even the ones that aren’t hitting that threshold sitting at like 70 something percent, it’s like still too low there, but there’s only a handful that I really needed to correct. And they’re like the trickiest questions of the bunch. so I’m trying to finalize that now. but even as I was going through, you know, cause our cadence will be okay. I build a prompt, we test it against a
various images or we already know the answers, okay, how does the LLM perform with that in mind? And then I go through the rationale for all the wrong answers, plug the prompt back in and figure out, okay, how can we improve this to get the right answer going forward? And so as I was doing a lot of this today, some of these incorrect answers, they were just flat out mislabeled, like, you know, where,
property was supposed to identify a driveway within the subject property. was mislabeled as a no instead of a yes. Like that was just flat out wrong within our own data record keeping. So those were easy fixes to make, but like the AI model is actually correct. Like it had the right rationale. just, we had the wrong answers. And in a similar situation, there were some images where
You know, my initial perception was okay to answer, you know, no for this answer. So we had like a, for example, a very heavily wooded property. didn’t think there were any structural improvements on it. so I answered, okay, you know, the AI model should say no. and, the model came back as a guess and he’s like, okay, yeah, there seems to be something in this Northwest corner hidden behind some trees here. then I’m like,
up this and so that I more carefully looked at the image the test image there like actually that is a more conservative approach there there does appear to be some man-made subject I can’t quite tell whether it’s a shed or a storage container but it seems like it could be so they I was better at identifying a feature than I was upon that pass
And there were few other examples like that to where, and I think I’ve commented on this in the pod more recently, but I just, day by day, I keep getting blown away by this. Like the fact that it’s this good already and keep in mind too, like it’s one thing to use LLM models just to identify various images, but you know, with land price, we’re trying to do it at scale and very cost efficiently. So in order to do that, we need to use weaker non-reasoning models. Otherwise, you know, the business model completely get destroyed.
spending hundreds, thousands of dollars on tokens to review properties that we’d be having to eat the cost of. So I’m trying to keep the cost per full review as efficient and low cost as possible, which is very tricky to pull off. So we have to naturally use weaker models to get this done. And the fact that our prompts are still getting such high quality results is just…
really mind blowing as far as okay, how good this accuracy is. And that as the models keep getting better and the, currently best models become cheaper over time, we can improve our results even more by incorporating more reasoning as well. so like this to me was just telling like, okay, if these models are that this good already, sure. They’re still making some errors. I’m correcting, but the fact that it is correcting some of my own responses.
based on its rationale and where I have to go back to the image and like, wait, did I miss something here? And, you know, forcing me to think more critically on what the AI was looking at. Like it’s a really good partner in that type of setup. Again, the fact that it’s already this good and, you know, we think, you know, at Serious Land Capital, I would just underwrite day in, day out, put our image analysis review process up there with anyone else’s.
around the industry. mean, not spouting this out of nowhere. Been doing this for years and AI is like correcting me on a number of images. You know, just look six months from now, 12 months from now or beyond that. It’s probably going to be better from just a straight aerial imagery review perspective, better than all of the real estate investors, all of the land investors like
This is coming now. Um, and again, I’m, I’m speaking from a standpoint of like, think we’re basically top of the industry doing this. And I see it being better than us in a relative short order. Like it, it is just truly stunning. I, if you are not working with AI models in a day in day out basis, it’s, it’s hard to perceive how fast progress is moving here.
but because we’re using these all the time now, I’m just, you know, more and more just like trying to hold on and, and, you know, try to, try to figure out, okay, how can we utilize this to the best degree possible? And also realize like, yeah, we just have to be humble about this to where, you know, we were the best at doing this for a while. This technology is just getting better. so.
If it’s going to be better than us, how do we still maintain an advantage and utilize its capabilities as efficiently as possible within our own business and distribute it, distribute those capabilities in an efficient manner to the rest of the land industry as well. So that is our primary goal here. Now I can kind of view it a blessing in disguise why the first iteration of land price did not work out because it opened up this door to
this newer iteration where I think, okay, this is far more powerful and a direction I was not necessarily thinking about even a few months ago. But that’s just the way things go sometimes. You have to be open to allowing the current reality and current technological progress and your own capabilities to match and let creativity kind of flow through what
the world and the universe is presenting to you and be ready to mold that information and knowledge that is coming your way in the best way possible. you know, I’m just kind of viewing myself, my team is more of a vessel to craft this type of potential in a deliverable format for as many of you and investors as possible. So very, very exciting stuff.
Looking forward to sharing more, but hope you appreciate this update in the meantime. With that, serious land dot capital for any of funding needs. Zero cost review of your land deals at land daily diligence. Facebook group just got off a live session about half hour ago. It’s very interesting deals there. If you want to catch the replay, some massive deals actually. And.
landpricer.ai, which I just discussed for this podcast. Remember landonconference.com, $200 off your ticket using my last name for the promo code. Only a handful of spots left. It’s only two months away. Rapidly coming up here. Hope to see you there. Subscribe and share to this if you haven’t already. Talk to you tomorrow. Bye.
