Blake Houghton đ on Nostr: Yea, I plan to switch my main card to AMD once the 2080 no longer holds up. It's not ...
Yea, I plan to switch my main card to AMD once the 2080 no longer holds up. It's not my AI card tho, so it's not a headache to change.
I've been seeing some of newer research papers and they are starting to realize how to get the neurons in the network to start mimicking the behaviour of actual neurons. It's all fairly primitive right now, sure, but simulating neuroplasticity and neural choice conflict were recent papers published. Simple temporal tuned knowledge graphs will be common soon (like Graphiti) and will allow for better "learning" post training. I wonder if initial training may even be as important if this gets good enough.
I am not sure how well a Riech will be able to control this đ€ The attempts to prevent up-to-date uncensored variants of their models has utterly failed.
They will need to openly Crack down on open source completely before that happens, I think.
But you're argument about Generative AI is solid. It all fundementally relies on fuzzy logic and people expect it act with perfect logic. Great for generating art but utter garbage for almost all coding tasks I've given them đ they struggle with Math a lot too. It's honestly impressive the models give accurate information as often as they do.
I guess I am still on the fence about it's trajectory over the long term tho. Like when electricity was first proposed and most piewed viewed it as a cool parlor trick, I can't predict the innovative ways humans will utilize and integrate the tool into our lives. It could be as ubiquitous as electricity is 100 years from now.
Do I believe in the wacky claims it will bring equality or apotheosis? No, that's ridiculous.
I've been seeing some of newer research papers and they are starting to realize how to get the neurons in the network to start mimicking the behaviour of actual neurons. It's all fairly primitive right now, sure, but simulating neuroplasticity and neural choice conflict were recent papers published. Simple temporal tuned knowledge graphs will be common soon (like Graphiti) and will allow for better "learning" post training. I wonder if initial training may even be as important if this gets good enough.
I am not sure how well a Riech will be able to control this đ€ The attempts to prevent up-to-date uncensored variants of their models has utterly failed.
They will need to openly Crack down on open source completely before that happens, I think.
But you're argument about Generative AI is solid. It all fundementally relies on fuzzy logic and people expect it act with perfect logic. Great for generating art but utter garbage for almost all coding tasks I've given them đ they struggle with Math a lot too. It's honestly impressive the models give accurate information as often as they do.
I guess I am still on the fence about it's trajectory over the long term tho. Like when electricity was first proposed and most piewed viewed it as a cool parlor trick, I can't predict the innovative ways humans will utilize and integrate the tool into our lives. It could be as ubiquitous as electricity is 100 years from now.
Do I believe in the wacky claims it will bring equality or apotheosis? No, that's ridiculous.