AI secrete source! – Gingercatsoftware

So it’s interesting comparing AI’s. I’ve been hacking away and recently I got privateGPT to run locally on my own machine. It’s an old and under powered beast and an average query will take about 15 minutes to process. It uses up all the ram, and the cpu’s sit at about 98% utilisation. The fans turn on and when they stop howling it’s usually an indication that the machine has spat out an answer (I think I saw 42 at one point!).

So this training of the AI and quoting of sources is interesting. PrivateGPT will produce sources – but with that product I’ve also run into my first AI hallucination (basically where an AI trys too hard and brings up false info).

This all begs the question – considering much of the available human knowledge used to train AI came from all of humanity, shouldn’t AI companies be taxed at a very high rate and pay that money back into the community with say the aims of establishing an international sovereign wealth fund that would eventually go on to supporting / establishing a universal basic income for all of humanity (one can dream I know)?

A couple of the publicly accessible AI’s that I’ve been tinkering with will not quote sources and will not tell you much if anything with regards what they have been trained on. But there is now this very interesting development that the Guardian have reported on which is that Sarah Silverman is current sueing OpenAI and Meta claiming AI training infringed copyright

https://www.theguardian.com/technology/2023/jul/10/sarah-silverman-sues-openai-meta-copyright-infringement

I’m going to be holding on to my pop corn real tight as this works it’s way thru the courts, and the big wigs in silicon valley work out what to do and how to do it. I told you 2023 was going to be a very interesting year for AI.