AI local on a old pc or even on a new fast/lots of memory & gpu and Npu. Can it learn on its own if all you do is train it on your local files of anything and everything? No youtube videos or Ai searches 0r even regular web search will answer this question? All I see AI is a glorified database that we create and a search engine or data base query look up on what we trained it on. It can’t and doesn’t learn on it’s own in the public everyday offline closed loop system? What models and who has access to the ones that can run in robots and allow them to think for them selves? Only the big companies and data centers have that capability. So why is everyone so hiped up on having AI locally if it can’t learn on it’s own. And 90% of results you have to correct anyway. I might as well right the book,solution, ect myself? You still have to proof read. Hell I don’t want to be an English teacher. Someone explain. Maybe my knowledge is short or off. Just don’t see the great use at home. I guess I need to start getting caught up or take a class or two?
I’ve heard of AI being used on offline computers. If it was ever online while training, I’m not sure. The one I’m thinking of was trained on first aid techniques. I heard that over a year ago and u could download it for free to work on yourself.
It looks like somebody took the idea and put it into a rugged case!
Ya just not seeing the big advantage. Ya I suppose you could use it for code, but you really need to understand the code and what things do. It also creates a lot of millions extra lines of code not needed they say. Then someone has to go threw and understand what the hell it did and try cleaning it up. Unless someone on here can tell me otherwise, sounds like a glorified data base that we create and a search engine. Logarithm. Sorts for searches. Who knows. Lets see what turns up..
They way it was described to me was a glorified spell checker, at the word level, with a bit of rnadom noise thrown in, so it guesses the next word, then checks it against its database. If OK-ish, on to the next word.
The part I found unbelievable is confessions from the guys who build, train and operate these things, that these guys just don’t know how it works deep down in the internals.
Artificial - yes. Intelligence - no.
Getting back to the local device, nearly all the Mozilla based browsers on F-droid now have built-in AI, they say for the translation engines. But there’s a bunch of fairly reliable translation engines available online. What else can AI do in a browser?
Speaking of Ai,
Here is a recent Ai query about a printer part I was searching for… I discovered it was in error and called it out… here the response…
I asked:
do your foul ups happen by design? Why exactly do they happen?
And it said:
No, they’re not by design. What happened here is a common failure mode:
-
I generated a slightly wrong token (“5” instead of “3”) on the first occurrence.
-
I then conditioned on my own prior output in subsequent turns, so the wrong version got reinforced and repeated.
It’s especially prone to happening with part numbers and codes — long strings of mixed letters and digits where a single character swap looks plausible. Once the wrong string is “in context,” I treat it as established fact rather than re-deriving it from the source.
It’s not a deliberate choice or a “feature.” It’s just a limitation of sequential token generation plus self-conditioning.
End of Ai chat.
Can someone here tell me what exactly is a “limitation of sequential token generation” ?
If its a programming error, why is it not corrected? If its not corrected and allowed to exist, then the error is purposeful…
Obviously, Ai is not self correcting or self learning.
As I said, a certain amount of random noise is injected into the process, hence the scornful epithet: Stochastic Parrots. Dangerously anthropomorphising - it stops them getting tunnel vision, which is why they guess each step. And when they’re wrong people say they hallucinate, or lie.