yes, I'm claiming that the NV-AI hype bubble will pop (which almost everyone expects to one degree or other).

but flash endurance isn't a strong argument here. you probably have O(TB) of flash, and aren't going to produce PB of swap writes any time soon. if you do a lot of swapping to a small flash device, it'll happen sooner.

I'm typing from a quite old 4GB laptop, which swaps heavily to a 250G SATA ssd. sure, it's not great, but it also costs zero. currently 9GB of swap is used, and it's not really noticeable. if I open 20 more tabs, it can introduce pauses.

google says this drive was released in 2014, and SMART says POH is about 10 years.

SMART also says wear leveling count is 665 and total written is 165327189538 LBAs (78834 GiB, or 338 drive-writes). I'm not expecting it to die soon, though using a 4G laptop is a bit of a stunt these days...

the point is that a system that has sustained heavy swapping for years has not generates so many writes to worry much. a modern system with 10x speed and 10x capacity (and probably less RAM deficit) would have even less effect. even for QDR with it's few-hundred cycle endurance spec...

AI is not a plagiarism engine. It can be used that way, but is not inherently so. It is not necessary that a trained LLM be able to faithfully reproduce every document in its training set. The entire structure of an LLM is not storage, but at least in principle, generalization: extraction of a somewhat abstracted "structure" of semantically similar "concepts".

But we also need to talk about authors' "rights". It's well-established that reproducing a work is infringement. There is a lot of caselaw about how much may be reproduced without infringement. But the idea that an author should be consulted before ANY automated use of their published (public) text? No, just no.

that is, reality exists and consciousness is "painted" on top of that?

IMO, anti-materialists are merely uncomfortable with the degree to which they understand neuroscience and related topics (including, btw, capabilities and limits of LLMs). Chalmers, for instance, basically insists that the Hard Problem is Hard simply because he finds it hard.

the reality-is-illusion meme is self-consistent (panpsychism, simulationism, dream-of-god-ism, whatever). merely being self-consistent isn't good enough.

the alternative (and there is only one) is physicalism and its epistemology, science. the main appeal of this is parsimony, often referred to as Occam's Razor.

I'm always mystified why consciousness is so often claimed to be undefinable.

it's not that hard to hit 300 cycles on flash.

there is no sign of any impending breakthroughs that would change flash economics much.

slc-mlc-tlc-qlc was very nice but plc will not happen. layer-based flash was also nice but it is ultimately linear (more layers, more cost, lower yield). dimensional shrinks are already stalled because of a tragic electron shortage (per cell).

I guess there's no harm in pining for some other NVRAM technology (spins, etc).

what happens is that the industry goes through certain discrete technical upgrades. for instance, EUV in fabs, or GMR disk heads. none of these are really planned, none of them are exponential. and they usually interact with other phenomena (such as Dennard scaling).

in a sense, the phenomenon is more like "expectations are exponential, and this motivates manufacturers to schedule updates".

hard disks are still improving, arguably similar to how they have in the past, but there are limits to demand. the consumer market has mostly dropped out, for instance due to flash.

even in flash, there is no exponential scaling in devices. people got excited in the initial startup, when for instance, mature TLC is so much better than early SLC. but all that's over: it's both mature and we'll probably never see PLC. even QLC is interesting in that it illustrates that most of our storage is very cold.

mystifying to me that shared, multi-user machines are not thought of. for instance, I administer a system with 27k users - people who can login. even if only 1/10,000 of them are curious/malicious/compromised, we (Canadian national research HPC systems) are at risk. yes, this is somewhat uncommon these days, when shell access is not the norm.

but consider the very common sort of shared hosting environment: they typically provide something like plesk to interface to shared machines with no particular isolation. can you (as a website owner or 0wner) convince wordpress/etc to drop and execute a script? yep.

but the basic premise (shared among a lot of ai-doomers and ai-shamers), is that the bigs have somehow raped society (by training on everything available). this needs to be challenged: it implies quite a strong model of IP ownership, which is not what appears in law, or in founding documents (which are quite different from current law).

Considering that most distros are basically just a new set of desktop backgrounds, this seems like a sure thing!

Ubuntu differs from Fedora only in newbie stuff, for instance.

presumably also advice from them.

but the idea is dubious: writing/reading is a different transaction than speaking/listening.

after all, most routers/WAP/gateways that you buy today will have linux on the inside, configured similarly.