Path: csiph.com!eternal-september.org!feeder.eternal-september.org!nntp.eternal-september.org!.POSTED!not-for-mail From: John Ames Newsgroups: comp.os.linux.misc,alt.folklore.computers Subject: Re: naughty C (was Re: naughty Pascal) Date: Mon, 12 Jan 2026 09:14:56 -0800 Organization: A noiseless patient Spider Lines: 70 Message-ID: <20260112091456.00001487@gmail.com> References: <10jn4al$15pib$2@dont-email.me> <10judtj$3eijl$2@dont-email.me> <10juqpa$3i6m5$1@dont-email.me> <20260110184115.4782eb44@coppelia.commodorejohn.com> <20260110233432.503028ee@coppelia.commodorejohn.com> <6963b9c5$0$31894$426a74cc@news.free.fr> <10k1n30$34ilb$1@paganini.bofh.team> MIME-Version: 1.0 Content-Type: text/plain; charset=US-ASCII Content-Transfer-Encoding: 7bit Injection-Date: Mon, 12 Jan 2026 17:15:02 +0000 (UTC) Injection-Info: dont-email.me; posting-host="1381e039aae46fb34971eb0d7574b47f"; logging-data="2566049"; mail-complaints-to="abuse@eternal-september.org"; posting-account="U2FsdGVkX19Xz8IAnZiJ5quVG+Ye70MWaHTIs6GhwZM=" Cancel-Lock: sha1:v4mqy9CNnKA97yOK2pD1eNXp40k= X-Newsreader: Claws Mail 4.3.0 (GTK 3.24.42; x86_64-w64-mingw32) Xref: csiph.com comp.os.linux.misc:80996 alt.folklore.computers:233567 On Mon, 12 Jan 2026 02:42:42 -0000 (UTC) antispam@fricas.org (Waldek Hebisch) wrote: > Now, concerning burst: AFAIK AI companies use investment money to > cover cost of operation (whole or in significant part). If there > is burst, they will have to stop operating literally closing > their datacenters. Basically only things that generate profits > or possibly some research by companies that have other sources of > income and still want to continue research. But that would be > at much lower scale than currently. This is the key point on which the "but the tech will remain!" argument founders. It's not just that research will stop being funded, but that it costs so much just to *run* that the only thing keeping the lights on is regular infusions of billions in investor cash - and, in OpenAI's case, generous subsidies on raw compute from M$. Once *that* dries up, they're either going to have to magically become orders of magnitude more efficient (which they've showed *absolutely* no aptitude for - and China had a good laugh pantsing our tech sector on that score last spring, but it's sounding post-fact like their gains were, um, not as impressive as they'dve liked everyone to think) or price to reflect the *real* cost of operations - and if they can only get a fraction of their userbase to pay $200/mo. for a Magical Chatbot Friend, good freakin' luck squeezing any more blood from *that* turnip. The much likelier scenario is that when the funding stops flowing in and the business plan fails to advance from "???" to "profit!" things will just...*stop.* Obviously, it'll still be *possible* to write this kind of software and run it in a slower, more limited context on stock hardware (high-end GPUs will be going *real* cheap for a while, but they also cost so much to run and require so much infrastructure that it's questionable how cost-effective it'll be to reuse them,) and possibly the tech will find its niche in certain fields (as "expert systems" did after the first "AI winter,") but the economics of doing "generative AI" at global scale are simply *insane,* and the current offerings in their present form simply will not survive once taken off VC life support. (Again, I'll recommend Ed Zitron on this, who has written *extensively* about it over the past couple years, and has done a lot to break down how much it actually *costs* - https://www.wheresyoured.at/oai_docs/ ) > But there were attempts to replace a lot of professionals, > notably programmers. Examples indicate that "AI" can create > small, trival pieces of code but does not really work for > bigger and more complex things. To be useful for programming "AI" > and way it is used must be significantly improved. It is possible > that slower, gradual improvement will lead to "useful AI". > But it is also possible that alternative approaches, currently > underfunded due to AI race, will progress and be used insted > of "AI". > > [...] > > So, it looks that for general AI we are missing something > important. For applications ANN apparently struggle with > tasks that have easy algorithmic solution. So natural way > forward with applications seem to be via hybrid approaches. > But AI crowd seem to prefer pure ANN solutions and tries > to brute-force problems using more compute power. Exactly. It's entirely possible that programming twenty years from now will look very different from programming today, and there *may* come a point where we "crack" using ML techniques for software development in a productive, reliable way - but it's *not* gonna come from the current crop of clowns, nor through burning infinite money on throwing *MOAR COMPUTE!!!1one* at the same stupid "solution."