I mean I imagine an LLM is able to generate more entropy from the sheer computing power put into it, but I agree traditional digital stenography methods are MUCH more cost/power efficient than an LLM.

(Not even to mention the amount of cyclic redundancy youd probably need just to get a message across)

Depending on the environment I suppose texr-based could be beneficial vs (relatively) large media files

I could honestly see this as one of the few legit uses for LLMs.

Throw some data ino it and make some “natural text” to obfuscate it.

Source: 🤷‍♂️ trust me bro

Dunno what else to tell ya cause they are moving to open source, but hey googles free if you want to find out for yourself

4% of US alone is 12 million people.

If even 25% of them decide hardware purchases based on driver support, 3 million sales isn’t ignorable.

(The number of PCs sold globally per year is similarly 300,000,000, so even then theyd lose out on 12 million potential sales YEARLY)

The market is also pretty shit post-covid, so I’m sure every hardware company is dying for a way to boost sales metrics.

With the linux server market share and recent ai boom, theyd have to be more than just blind deaf and dumb to not release linux drivers.

Maybe this was true back in like the early 2000’s?