Complete computer DWEEB with a question about A.I...!!

ulrichburke

ulrichburke

ulrichburke

Associate
Joined
9 Sep 2018
Posts
6
Dear Anyone.

It's just - I'm a programmer. Not a good one, but I like doing it! And most A.I. things seem to be written in Python, which is a language that's about 20 years old, give or take.

So how come people have only just discovered they can write A.I. with this 20 year old language? I mean if it was a new, custom designed language then yeah, I'd understand it. But Python!?! Why haven't people been doing A.I. with it before? I mean if it can do it now, it must've been able to do it for the last 20 years or so cos it's the same language. When did this "Wow!! We can write A.I. with this thing!!" moment happen?

And also - again as it's a 20 year old language it's got to be nothing NEW, just an algorithm. So SOMEONE'S taken this old language, found a new algorithm to write using it and is marketing the basic algorithm as A.I., am I right? Who's discovered this and why did nobody realise Python could do this before?

I'm not intelligent enough to USE Python that well - I wish! - I'm just fascinated in the above questions.

Yours puzzledly,

Chris.
 
If it wasn't for your join date I'd say this written by AI to try getting engagement
 
It's not a language question, you could use many languages.

Small weighted models have been used for 20+ years, though these were not LLM, more statistical models you trained with results of a process and they predicted a configuration or formulation based on input material QC results. We used them to predict ratios of polymer mixed to get dissolution results in specification. The actual QC results were then trained back to the model in batches to improve accuracy.

What we didn't have was the hyper scale high bandwidth compute required to hold a large model in memory and train with billions of tokens at a useful rate hence LLM are more recent.
 
Dear Decto.

Thanks, that really helps me understand - no sarcasm! It's just when I saw A.I. being pushed in all directions from nowhere my interest in programming kicked in - like I say I'm not a GOOD programmer but I've programmed in various versions of BASIC, Fortran and Cobol for years - and I wondered what these new programs were written in, expecting a brand new language with brand new capabilities. Much to my surprise, Google said a popular choice was Python which I knew was about 20 years old, even though I didn't know much about it and had never used it. So it piqued my interest as to how come it was SUDDENLY being used for A.I. and nobody had thought of this before. I didn't know it had been used on a smaller scale before - thanks, Decto - I just thought that as a 20 year old language was being used for it, AI must be more of an algorithmic technique than something totally new.

And reducing it in simplicity to something a brain my size - anybody seen the magnifying glass!?! - can understand, it seems to be a mass nest of Repeat-Until loops going over and over data until All the Bits Fit. That even explains how it hallucinates - it could quite conceivably find a way All the Bits Fit to its loops that has nothing to do with Reality, because it doesn't really understand what Reality IS. If all the bits fit, it assumes they fit reality. Telling it if it's right or wrong is our job - if the output's hallucinogenic we just shove the data back in till it comes up with something that fits all its loops AND fits reality!

It's amazing what can be done with a megafast system and a few thousand (hundred thousand? What do I know!?!) Repeat-Untils and While-Wends, isn't it!?! And of course when it gets a basic something right, I'm guessing that goes into a database so it can refer to the database and not have to send that particular lump of data through its loops again. And refer the new lumps of data that have just gone through its loops to the lump it knows is right to see if the new lumps agree with it. IF they do, THEN add new result to database and get the next lump of data. Can see how it's doing it now (I think!)

Thanks again, Decto!

Yours respectfully,

Chris.
 
Dear Soldato.

I've got Asperger's so I have difficulty in telling when a neurotypical's being serious or not - but I assure you I'm not a robot. Brain damaged yes, robot no. You tell me how you'd like me to prove my humanity and I'll do it (unless it involves money - I'm skint!)

Yours respectfully,

Chris.
 
You must log in or register to reply here.
Back
Top Bottom