Only if you let it. And complexity is true for any codebase as it grows (even in pre-AI world) which is why unit and integration tests are vital.
It wrote some rather dense unit and integration tests. The Unit tests were done in pure SWI Prolog, and the integration tests were done with bash scripts.
It was interesting that ChatGPT already knows my RCA method and treats it as a USPAP-compatible approach!
It is interesting that it is so fluent in Prolog - an old favorite of mine. I once wrote a Yacc++ interpreter for a variant of Red Pepper's Manufacturing Scheduling Engine (a C++ implementation of NASA's Lisp scheduling engine) back around 1994-1996.
Prolog is a "Logic Language" that proves the code to be true. So, ChatGPT parsed and analyzed USPAP and wrote programs to validate a submitted appraisal's data (including the appraiser's qualifications) and report. This is important for implementing standards and protocols in future systems. Society, i.e. people, need to know that the code systems implementing their laws are indeed correct. Prolog is the type of language suited for this purpose.
So, understand: Yesterday evening, ChatGPT and I completed a comprehensive system to automatically audit an appraisal, ensuring it is fully USPAP compliant. The two of us created code that I have already checked into my GitHub repository. BTW, GitHub is often used by large and small AI companies to store code for distribution to their customers. Habib, who founded Writer.com, used GitHub in its initial years to distribute her code and updates to major customers.
It has been checked into my GitHub repository
https://github.com/wcraytor/NewUSPAP as a Private repository. I may make it public after some testing and more documentation writing today.
Invariably, what GitHub generates may run just fine on the initial run. However, it is NOT an appraiser, and an appraiser needs to go over its logic and fine-tune the code. An example was code that specified which approaches were used for appraisal. It had at one point simply "have_sales_cost_income=TRUE" which means that the appraisal has all three approaches to value. I had to inform it that only some of these values might be needed, so the code was changed to something like "has_sales_cost_income(<appraisal id>, [sales, income]), plus some other associated changes.
The code is written in R and Prolog. R can directly execute Prolog, more or less, by generating the Prolog to scripts and then executing those scripts in SWI Prolog.
Caveat: Prolog is a very different and challenging language to learn. In fact, so difficult that after an initial splurge of Prolog code around the end of the 90's and early 2000's, it went downhill in the US. It is still used in Europe. It originated in France. Sweden still maintains and sells SCIStus Prolog - one of the major implementations. There are many versions still used:
en.wikipedia.org
Anyway, I have worked with it on and off over the past 30 years. However, I assure you, there are few out there now using Prolog for appraisal, if there are any who can use it at all. It will be a new language for most developers in this field, which will likely be needed for protocol implementation in valuation and other fields related to the automation of regulations.
Once I have tested it and am happy with the initial version, I will likely publish it, i.e., make it public.
It is to me so interesting that ChatGPT can instantly analyze USPAP, reformulate it in a more concise version, and present it as a "Proposal" to avoid the Copyright protection that USPAP has. Yes, ChatGPT has its own built-in attorney. Then it also was able to implement my request to use SWI Prolog and R to provide software capable of auditing an appraisal to determine if it is USPAP compliant.
Not only that, - it created a "California BREA" overlay to determine if it also complied with California BREA requirements!
So, all this in several hours of work. Think about this hard. This is REVOLUTIONARY.
