I was compiling nutrients on diet plans yesterday -previous results have been good, but I switched LLMs for a trial run- it was off by a factor of 10 for cauliflower rice. They are great for some grunt tasks, though, like converting blobs of text data into re-ordered csv files.
Are they though? Did you compare against conventional methods to check it didn’t change some of the values? Because it did when I tried to use it for a grunt formatting task. Now I don’t even trust them to literally copy and paste text without introducing subtle changes.
>compare Yes, in the csv use-case I was scraping, sorting, and moving published nutrient data for ready-made meals into a spreadsheet. The site was scraping was very difficult to navigate so I was forced to do this row-by-row. It was easy to proofread in that situation.
I've also found certain ones to be good at generating weekly specialized diet plans. One I've used is OK at medical symptoms & treatments --but the prompts were very detailed & focused on single chronic illnesses. That said, I fully agree that you cannot trust the 'retail' LLMs left on their own.
Everybody keeps telling me that these things are great for things that literally never come up in my life