@retipsterseth In Oregon, the Residential Specialty Code (ORSC) defines presumptive load bearing values certain types of soil (https://codes.iccsafe.org/s/ORRC2023P1/chapter-4-foundations/ORRC2023P1-Pt03-Ch04-SecR401.4). I’ve found that ChatGPT does a pretty good job if I can take that NRCS report and point it directly at this code and ask it to compare the report soil types against what is defined in the code. It also does a pretty good job of flagging poorly draining soils and expansive soils based on the soils types. I use this code for comparing soils in WA, OR and ID, as it’s a good litmus test, even thought it is an OR specific code.
Regarding AI for due diligence, I use it as a tool for every review that I do, but its very much a trust be verify game. As you’ve experienced, it’s definitely lead me astray, but answered very confidently. More often then not, it will extrapolate information that is not codified. It will grab info from public hearing notes, articles, announcements or draft code, then state it as fact.
I find it very useful when I provide it a code section or chapter and ask it to extract specific data, such as lot coverage and building setbacks. When I get into less defined areas, such as stormwater and erosion control, I’ll ask it to cite the specific code it is referencing. I’ve had several instances where it has corrected itself when I’ve asked it to cite “currently adopted” code for a specific jurisdiction.
I also use it a lot for analyzing all the small print in plat maps and CCR’s.
AI is a great tool. But it definitely takes intuition and expertise to wrangle it.