[Gmix]

I’ve recently been using Ai to help with a project
It’s been remarkably wrong on various occasions even when challenged it will often say it’s right until you point out the flaw in the information or explain why it’s wrong, then it will say your right to challenge me & admit it was wrong & explain why ; it was because of xyz and in future it will check & not do it from memory
Even when it does that it still gets so much wrong it’s a very unreliable tool.
If your using it for your grow I would definitely recommend you don’t rely on that as a tool alone and get an opinion from someone that knows/grows

 

[Gmix]

I set rules for it to follow
I asked how many it had broken; all of them

 

[gullywompr]

I’ve recently been using Ai to help with a project
It’s been remarkably wrong on various occasions even when challenged it will often say it’s right until you point out the flaw in the information or explain why it’s wrong, then it will say your right to challenge me & admit it was wrong & explain why ; it was because of xyz and in future it will check & not do it from memory
Even when it does that it still gets so much wrong it’s a very unreliable tool.
If your using it for your grow I would definitely recommend you don’t rely on that as a tool alone and get an opinion from someone that knows/grows

I run a local LLM that I've added a large quantity of published scientific research on cannabis specifically (much of it textbook level), and programmed it not to give answers if there's no reference behind it, i.e. it cannot answer with opinions or broscience, it can only give credentialed answers, i.e. it's reliable (unless you prompt it with some misinformation, then of course it can be incorrect in it's details). It's tricky to find one small enough to run on your machine, but still with enough parameters to give meaningful responses. I've gotten 4 billion param models to run on a raspberry pi, that takes 16GB. But yeah, that makes a HUGE difference in reliability.