Indeed, Ollama made it easy to download n test these LLM's.Thanks, Ollama.
So easy in fact my new 450GB SSD only has <100GB left
Just glad I have Fibre to the home.
Seen a few graphs of GPT-3.5 v GPT-4 after the question is asked a few more times.
GPT-3.5 improves faster and gets a more correct answer after about 3-4 cycles.
Stable Diffusion XL has the same sort of convergence, landscapes are the best after 3 steps.
Zero step is where an AI gets the correct answer the first time.
At the moment these LLM's are not on all the time and it is back to zero when power goes off.
When memory is added for historical recording, perhaps things will get a big jump in capabilities.
Long Term Memory will be very handy for a personal AI assistant, I don't need a bot that has worse memory than me.
Add a bit of self learning capability and things get interesting.
Statistics: Posted by Gavinmc42 — Wed Mar 27, 2024 6:13 am