You are not logged in.
- Topics: Active | Unanswered
#1 Yesterday 19:20:18
- chomwitt
- Member
- Registered: 2019-09-24
- Posts: 236
bigLLM owes a big thanks to FSF. And FSF owes thanks to bigLLM ?
alt title: choosing food (data training diet for your LLM)
Funny thinking the interplay of bigLLM and libresoftware.
Vibe coding is one central sales pitch of bigLLM . Obviously .. most libresoftware is free (as in no price!) .
So uploading all that code to LLM big data centers is :
- a fast and cheap way to create a good pitch that it will attract the usual attention of the usual big-tech (as in big pockets) lovers.
- a fast way to give credit to your ultra hype balloon in order to start tapping on investors money and politicians egos.
So as a hypothetical LLM investor that is on the LLM bandwagon tapping on piles of public subsidies , water , land , and investors money i should first of all thank RMS and FSF ??
What do i miss ?
Also i wonder if bigLLMs (as in big in size and belonging to big pockets) are ok with binary files ? Could i give them e.g a closed nvidia driver or some nvidia blobs and do vibe coding with it ? I wonder what policy nvidia (the ex-gaming-gpu corp and currently most big AI company) has on that issue ? I wonder how Nvidia could protect it self if a local LLM could cracks its blobs . I wonder if Linus would approve that vibe coding since he poses as a show me the code guy. (blob you say ? ....xmm.. that piece of '''software''' that is even less intelligible that machine code... signals calibrating on semiconductor-level unique quirks ... like speaking a language no one understands? no ISA involded? interesting ... so maybe an LLM would throw up a blob??)
And generaly i guess bigLLM want good and cheap training food. In an cut-throat-evolutionary view of the world imagining big trucks offloading big piles of old books (1) and libresoftware in front of big LLMs for training through destructive data-decomposing kind of aligns perfectly with that world view .. dont you think ?
Speaking of bigLLM data diet .(or web-crawelrs's d-diet) . I wonder.. is there a data that cant be eaten ? That would make it throw up ? Or bad for digestion ? Or totally un-eatable ? And how that is related to the essence of what is 'software' and what is digitized data .
I've read recently many news about microsoft increasing the window security patches. That means that it's binary is increasingly used as LLM food . But LLM controlled by what would ms consider hostile. So it's kind of peculiar funny irony what happens. How ms or other big companies would protect their binaries from hostile LLMs? I wonder if for a LLM binary-assembly-higher source code structure is rather irrelevant to it's abilities. And if yes.. why not make windows open source ? Why ? As a defense maybe . Or because it's irrelevant. And if source is irrelevant what is open source ? Maybe a defence is windows as firmware on a chip ?
..So from an LLM's perspective both Linux and Windows are very good structured data food with Linux perhaps been a little lighter to it's ram-guts due to its higher source code ? (like a sauce on a food ?)
(1)@Inside the Warehouse Where Amazon Scans and Destroys Books for AI Training
Last edited by chomwitt (Today 07:34:08)
Online
