Ask HN: What's stopping a hardwired AI
Simple question: can't AI exist as a "hardcoded" / coded purely using transistors and logic gates with builtin weights? (Of course from an open source model) And then just power it so every output is PHYSICALLY generated based on the input. (For example a binary added can take in information, add the bits and output with very less power consumption, can we do the same for AI?) Remove the obvious stuff like how AI isn't perfect yet ..
You were thinking about this? https://taalas.com/
Yes
… thanks, I was looking for the link for XRumer everywhere
It will be obsolete on day one, for one thing.
Yes , but I think the power consumption being 10x less is better here and don't forget way less memory
Well there are some cases where extreme low power consumption is everything. Say wakeword detection in something like Alexa but battery powered.
in terms of current event knowledge, perhaps. but in terms of coding and system admin, the current frontier model weights could easily be cast into silicon and be useful to every system it is installed into. every new computer comes capable of writing all its own software and doing its own expert system admin. that will definitely be a thing in the future.