3 ms·
TBH I'd very much like to have an "HAL" that contains a world of knowledge (even if it never updates, or only gets updated once few years) at home. I wonder wh
by ferguess_k 2y ago
TBH I'd very much like to have an "HAL" that contains a world of knowledge (even if it never updates, or only gets updated once few years) at home.
I wonder whether the current technology is feasible for a replication of HAL.
- Very good chess player -> Check (And probably can be a top notch Go player too, although I don't play chess or go)
- Very good knowledge of history, science, technology, everything -> I'm not sure if it's possible. And I'd assume we need to send scans of all books, because for example it should know many fine details and educated guesses about a certain historical fact.
- Understand human language pretty well -> I actually think ChatGPT is good enough if I choose my wording clearly. But voice recognition might prose some challenges. It also depends on how good the microphone is.
- Have pretty good voice generation. I'd like to choose the voice too! -> I guess it's OK-ish nowadays? I listened to a few AI generated clips and they are pretty good, not sure how practical it is, though.
Anything else I missed? I know HAL also has its multi-decade objectives in mind, so this is different from the LLMs which don't seem to have a very long term of memory.
- WillAdams 2y agoAll I want is an LLM front-end which will run a (tested/working) prompt[1] on each file in a directory, and collect the output into a single file/response at the end which could then be run on the files in that folder to get them renamed. 1 - for the pdf file, create a move command for a batch file which will rename the file using $<check amount>_<date from check in YYYY-MM-DD format>_<invoice id>_<PWS ID>_<name of company from Population line>.pdf as the filename.
- morkalork 2y agoBasically the computer from startrek right?
- ferguess_k 2y agoThe one from 2001/2061, but yeah that one is also good.
- thorum 2y ago> Very good knowledge of history, science, technology, everything All of English Wikipedia is about 100 GB. A local LLM that can run searches against a local Kiwix server would work. Local LLMs can’t really do this correctly yet, but it’s not inconceivable in 1-2 years.
- ferguess_k 2y agoWhat I'm thinking is a bit more than that. E.g. it should consume all books in the world for topics I'm interested in. Wikipedia is too shallow for in-depth discussions but it's a good "appetizer".