I have used AI to rephrase a paragraph or letter and the results have been great 95% of the time. The results is sometimes too formal or uses words I wouldn’t use.
For structural engineering questions or solutions to problems I give it, AI does pretty good. However it makes mistakes based on improper assumptions or failure on my part not being specific enough. I typically ask it to include reference material it used in its calculations. Just keep in mind your talking to a five year old and have to explain everything in detail.
I can’t see having an artificial memory bank/persona parrot information that may or may not be pertinent. Why not just use your own words to report what you observe; and recommend that a specialist be consulted, if you reason that to be necessary?
AI is continually learning from us. I did a short session with Grok and had to correct it several times on key issues.
Reports I’ve heard indicate AI’s are learning form each other too. Also that AI in general is decent at math related fact type information, but is biased on most social or political issues.
I asked Al (technical) questions I already knew the answers too. In many cases it’s answers were incorrect. When presented with the facts (and references that back the facts) Al thanked me for the correct information. I would definitely old school verify anything that Al tells us.
I wonder how many scams are ongoing throughout the internet related to Ai. I am right now battling a company called Coursiv to keep them from fraudulently debiting my bank account on a monthly basis. In early July, I signed up for a promotion that briefly introduces you to various models of Ai. The cost was $19.98. Immediately upon remitting the funds, the same amount was debited from locations in two other countries, including Cyprus. I reported it to my bank in a formal dispute and I am still awaiting the outcome. In the meantime, I just got another debit from Coursiv in Dover, Delaware for twice the amount: $39.98. There was never any mention of recurring installments. It seems like another case of turning Ai loose to scam the public; and then blaming it on Ai when you get caught.
I’d be curious to hear your experience with the output given some iterative learning of the model. Rephrasing a prompt once or twice can give much better results. It takes a bit of back and forth it seems, but you can prompt the ai to write the response in “your voice” or a particular style. While some people say it’s like talking to a four year old, I would say it’s closer to a fresh out from college. They are fairly smart and have access to almost limitless information, but may need some guidance the first couple times you give them a new task.
