Clickforms 3.6 UAD

[RCA]

April I believe. If it works as they claim, it will be superior to others.

I have no idea why the GSEs are willing to trust AI filling out a form based on wildly inaccurate and random MLS data, or lidar sketches that are typically 5-10% incorrect, but if that's what they want, that's what they'll get.

You have to understand the nuances, ref. AO-41: If you document your prompt to AI (i.e. your request for info) the exact response, THEN it is supposedly considered nothing more than data. That is to say as long as you have documented your exact request and the exact response, there is nothing per se wrong with using AI responses. HOWEVER, whether the appraiser should accept that data is wholly on the appraiser. If the AI system misunderstands the context of the question or simply misunderstands the context for some other reason (prior questions) or is simply incorrect, then that is something that is the responsibility of the appraiser to catch.

So subtle it is: If you get a response from AI that supports your existing value conclusion, you can certainly mention that in the report as support. But, understand, it would be reckless to depend on that "data" for a given value conclusion if that is all you have for support.

=====
A smarter move is to run MARS or GLMNET regression on a data set, provide the parameters (with some reasoning) and permanently store a copy of the data used as an excel or sqlite file, as well as the script, so that if pulled before the board or you wind up in court, you can rerun the regression to prove the model you used was objectively obtained. That would do a very good job of protecting you from an accusation of bias or incompetence.

=====
Again, a subtle point: As long as it is well-documented, there is nothing wrong with using AI responses to add support to your value conclusion, but you can't just blindly accept the results on no other basis.

 

[RCA]

And by the way for most people asking about the value of some house, the AI will likely just query the web and combine the results using its training algorithms. If you go to the trouble of uploading MLS data sets and other data, - and if it knows you are an appraiser, it will likely do much more work. But it won't lie: It's going to say it is totally dependent the accuracy of your data and there are all these things A,B,C,X,Y and Z it doesn't know. So, the response, at least from Claude, could be very conditional and probably not worth that much.

However, if you do a very good job of making the prompt, telling it you want to have it do a MARS regression on some market area data you have uploaded, and based on that, and using the Residual Constraint Approach, create a sales grid and value conclusion, well, you might possibly get that. I don't see why not. But you would have to complete the sales grid expanding upon what you think are the latent variables that make up the residual for each comparable and the subject. Oh and not forget, all the AI can do is give you a formula to help calculate the residual for the subject --- that is also something you have to do.

And then, if you don't know what the **** you are really doing, are you, as an appraiser, really that much more qualified than a homeowner to value their property? - Well of course, assuming you are a disinterested party, assuming of course.

 

[RCA]

I should also mention that Claude.AI can give appraisers instructions or even write R code for them to execute on their machines, provided they have R or RStudio installed.

If they are using Claude Code, rather than Claude, it can actually generate R code and execute programs directly on the appraiser's computer. Claude does have access to my open-source code, all CRAN source code, and my published docs on the RCA method. It could pretty much do your whole analysis and a good job of it, - provided you do an adequate job of monitoring it.

Which begs the question: What could a paid subscription really provide that ultra-high-IQ Claude Code could not? This question will only get more difficult to answer as the year progresses.

--- So, someone, in the end, may be wasting their money.

--- And I will, at least for another couple of years, be grinding away on advanced appraisal techniques.

What do you think of my latest https://journal.valuation-engineer.com?