Elliott
Elite Member
- Joined
- Apr 23, 2002
- Professional Status
- Certified General Appraiser
- State
- Oregon
An appraiser asked me a question, can you help them?
"If possible, if you have time, please try to explain for me, in layman's terms, how an entity such as ChatGPT can be familiar with the intimate details of esoteric issues like this. I became aware of the existence of AI very recently, amazed that ChatGPT is hand-walking me patiently through the process to address "Borland Database Exception Handling Errors" affecting every ACI report, for weeks, which ACI tech support has escalated to the ACI Development Team. Or like when ChatGPT tells me that "based on make, model, year, mileage, and symptoms," the parts store recommendation based on results of a diagnostic scan are not the right parts that I need. I'm not yet enamored of the concept to rely upon it to help resolve every unknown factor in life, although I can see how it might be tempting."
AI Response: "AI like ChatGPT gains familiarity with esoteric issues through training on enormous datasets scraped from the internet, books, forums, and other text sources, which include discussions on niche topics such as Borland database errors or automotive diagnostics. During training, the model learns patterns, associations between words, and sequences—like linking “make, model, year, mileage, symptoms” to specific part recommendations—by predicting the next word in billions of sentences repeatedly. This creates a vast web of statistical connections, allowing it to generate relevant responses without true human-like comprehension, much like autocomplete on steroids that draws from collective online knowledge.
Newer versions enhance this with “reasoning” capabilities, where the AI simulates step-by-step thinking by breaking problems into parts, testing approaches via trial-and-error (reinforcement learning), and refining based on rewards for correct outcomes, especially effective for technical fields like programming or math. For instance, it can troubleshoot ACI report errors by recalling similar patterns from developer forums or diagnose car parts by matching symptoms to documented cases across auto repair sites.
Appraisers benefit similarly; ChatGPT can suggest nondiscriminatory neighborhood analysis phrasing or UAD form tips drawn from public appraisal guidelines and discussions, but always verify outputs against USPAP standards, as AI lacks professional judgment. This training explains its patience in hand-walking through fixes that stump tech support—it’s recombining proven solutions from countless sources faster than humans search."
"If possible, if you have time, please try to explain for me, in layman's terms, how an entity such as ChatGPT can be familiar with the intimate details of esoteric issues like this. I became aware of the existence of AI very recently, amazed that ChatGPT is hand-walking me patiently through the process to address "Borland Database Exception Handling Errors" affecting every ACI report, for weeks, which ACI tech support has escalated to the ACI Development Team. Or like when ChatGPT tells me that "based on make, model, year, mileage, and symptoms," the parts store recommendation based on results of a diagnostic scan are not the right parts that I need. I'm not yet enamored of the concept to rely upon it to help resolve every unknown factor in life, although I can see how it might be tempting."
AI Response: "AI like ChatGPT gains familiarity with esoteric issues through training on enormous datasets scraped from the internet, books, forums, and other text sources, which include discussions on niche topics such as Borland database errors or automotive diagnostics. During training, the model learns patterns, associations between words, and sequences—like linking “make, model, year, mileage, symptoms” to specific part recommendations—by predicting the next word in billions of sentences repeatedly. This creates a vast web of statistical connections, allowing it to generate relevant responses without true human-like comprehension, much like autocomplete on steroids that draws from collective online knowledge.
Newer versions enhance this with “reasoning” capabilities, where the AI simulates step-by-step thinking by breaking problems into parts, testing approaches via trial-and-error (reinforcement learning), and refining based on rewards for correct outcomes, especially effective for technical fields like programming or math. For instance, it can troubleshoot ACI report errors by recalling similar patterns from developer forums or diagnose car parts by matching symptoms to documented cases across auto repair sites.
Appraisers benefit similarly; ChatGPT can suggest nondiscriminatory neighborhood analysis phrasing or UAD form tips drawn from public appraisal guidelines and discussions, but always verify outputs against USPAP standards, as AI lacks professional judgment. This training explains its patience in hand-walking through fixes that stump tech support—it’s recombining proven solutions from countless sources faster than humans search."
