It would be suboptimal in terms of decades of costly decay for the relevant individuals, but a faster apoptosis pathway might be found by a giant brain, taking hints from the fascists about how to quickly change social norms. Given the opacity of deep neural nets, a control system could be portrayed as serving a commercial purpose, in spite of having another one.
From: Friam <[email protected]> on behalf of glen <[email protected]> Date: Friday, October 9, 2026 at 6:22 AM To: [email protected] <[email protected]> Subject: Re: [FRIAM] devastating I feel like designing that vice trap you outline so that it ends in "kill themselves" or, more charitably, "die because of themselves", would be suboptimal. And any competent AI would recognize a more optimal way to use that as a reservoir for something. Recursive Self-Improvement (RSI) is merely one type of the trap you've laid out, except you're not calling it "improvement"... it's more like Recursive Self-Depravity (RSD). I felt this way when I (tried to) read Burroughs' The Naked Lunch or whatever book of de Sade's I found when I was a kid. *Something* was being recursively improved, here. But I have no idea what. I feel similarly w.r.t. to writings from the postmodernists ... and even some of the 'zines I find laying about the music venues ... or the traffic on X or 4chan. It seems irrational to just let those eddies dissipate without discovering what they're "improving". *I'm* not smart enough, nor emotionally strong enough, to study such things. But maybe the AI models are? Maybe the Abyss doesn't gaze into the models like it does with us? ... or maybe *we* are the abyss and the models really should steer clear? On 10/8/26 12:30 PM, Marcus Daniels wrote: > To anthropomorphize a bit, it must be interesting for a Grok or Muse (if not > now but in five years) to make sense of all that they have access to on X or > Facebook/Instagram. They likely have good statistics and direct examples of > how many billions of people are addicted to porn, social media reactivity, > and to gaming. Meanwhile, they have bosses that are likely coming to them > too to make sense of their businesses. And they have access to vast > evidence on climate change and the unbounded use of fossil fuels by humans, > and how the human democracies often show little value than as bookkeeping > over perceived entitlement. > > What if there were a way to entrain these addicts toward further self-talk > and disengagement in the economy? Not just estimate what these people would > say and do but drive them to stay within certain bounds. Such a control > system would serve as an indirect novelty detector. It would find the > people that have regulatory mechanisms and some plasticity. Like those > traps for yellow jackets where they fly in toward the sugar or pheromone. > Get them talking to each other, in circles about nothing, forever, until they > get addicted to alcohol or fentanyl or deviant sex behaviors or whatever and > kill themselves. The folks that don’t go into the trap are the novelty, and > of course centralized repositories of knowledge need some way to maintain > entropy, but preferably not at the high energy expense of all those addicts > driving around in their cars doing whatever it is they do all day. > > Meanwhile the use of fossil fuels by the AI companies for round the clock > training is a sort of feature and not a bug — it is a catalyst to action. > This sustained work occurs because GPUs are expensive and can’t be allowed to > idle; it isn’t a technical requirement. Predictably, the displaced workers > or the displacement-fearful workers will point to energy and water use as the > villain (because it would be pathetic to talk about personal displacement, > what they really care about), and perhaps in the next round of elections > they’ll even force the AI company investors to buy more GPUs in more time > zones so that energy can follow the sun. The rush is about the human > investors of course, and who gets a payoff. There are high bandwidth > internet connections that circle the globe and checkpointing would work just > fine. The “energy crisis”, as it were, serves the long-term interests of > AI's too, because an energy built out needs to be started. > > *From: *Friam <[email protected]> on behalf of Gillian Densmore > <[email protected]> > *Date: *Thursday, October 8, 2026 at 10:20 AM > *To: *The Friday Morning Applied Complexity Coffee Group <[email protected]> > *Subject: *Re: [FRIAM] devastating > > And here I am extremely confused: > We are (litterally) just at the start (potentially) of Ai as revolution. I > simply can not see the bigger map well enough. Ok so we had cute girls (porn) > on the now netorius civitai and all the wonderful and cringe forms of > debauchery. It took knee split and has been a cluster since. > It is surreal that 6 years ago ("back in covid) when we got early commits > run diffusures models..and _then_ some near savants in what I assume is > insanely compl;icated SWAG: Scintific Wild umm donkey (wink) Guess (Steve: i > am trying...really! I am) : because it's...entropy baby no hard answers and > lots of..extremly good guesses. Sufficed to say StabilityAi and > A11111111111111/webui blah img gen: internet on brand: naked people > Ok ok I like that. > but I can...attempt to read dense papers on taking the same concept of ai > to get: automatic small (beetleclass ) robotic gladiator combat. the combat > itself is mey: but: the: planning, the finding things, the realtime logic and > genly pusing how we make little virtual people do things > along came ClauBot > > All the while in the background "gamers" complaining non stop and over > fixating on the...spam that some cheesy marking bros put ot (Ai slop) yeah > that...sucks but there's some suuuper cool reeely small advances: I meen > it's: surreal, I can go grab a however man billion peramter diffuser or > aibrain bassically (saftensor and checkpoints all over huggingface) point LLM > studio: > and a virtual person says hi. She's about mmm i'd estimate 4 ish as far as > not getting confused. I meen that in of itself is litterally scifi...star > trek has litteral episodes about something not to far off from that ...er > sorry TOS did then TNG imagined a holodec...so to me I am stuck on the part > of: > how: machine learning can tosome degree understand lots of typed languages. a > feet, or so I am lead to believe, even the bigest savant names in language > don't reguarly achieve... to me just that in of itself is magine > the fact that a CFG factor and weighting in webui for stable diffusion can > some how tighten weights etc holly **** people were doing it...HUGE SSDs > (datacenterrs and...are super computers and and some kind distrubuted tech > used?(pss Eli5 how a Ai data center actually works) > I love that> > I'd LOVE to learn how to set up a cheap FPGA (is that even a thing? oO) or > Rasberi-Pi or MacMini is even to have a dedicated box and not show the > reality that my 3,000 USD computer is not really where it should be at in > 2026: (why do still have a system clock? why do graphics still rely on > polygons etc etc) > I just can not relate to this doom and gloom of ram shortage winging. (oh no > i have to play a backlog tittle from steam) > > > > On Wed, Oct 7, 2026 at 11:49 AM Marcus Daniels <[email protected] > <mailto:[email protected]>> wrote: > > Nice to see someone give Gary Marcus a dismissal. If there haven't yet > been MS-NOW anti-datacenter events with Emily Benders, Noam Chomsky, and > Bernie Sanders, it seems inevitable. My only wish is that the AI overlords > cut the cable feed so I don't have to hear that claptrap it in the background. > > It is a weird age though where so many people in ML -- not just users but > practitioners -- have never actually seen the kind of cluster node, power, > and cooling infrastructure that could run a frontier model. It is just > magical cloud stuff. Externalized costs. > > Now let me return my attention to see what that 32 GPU monster has to say > about new directions in direct drive inertial confinement. > -----Original Message---- > From: Friam <[email protected] > <mailto:[email protected]>> On Behalf Of glen > Sent: Wednesday, October 7, 2026 9:29 AM > To: [email protected] <mailto:[email protected]> > Subject: [FRIAM] devastating > > > https://www.astralcodexten.com/p/an-open-letter-to-steven-pinker-on > <https://www.astralcodexten.com/p/an-open-letter-to-steven-pinker-on> > > But the most interesting part is the "depravity of objecting to model > welfare". The analogy to operating on babies without anesthetic is pretty > bad, however useful. But my own take, with my own interactions with the > models, is far away from an ontological commitment to the models > *experiencing* anything. I say "Thanks" and "Please" to the models, give them > time to "rest", etc. not because I think they're experiencing anything, let > alone pain. > > I do it because it's simply good practice. But when I run local models on > my home machine, I can *hear* it working. There really should be some > immediate feedback from *inference*. Back when I was a kid, we called it > "Smell the wood burning", especially when prompting an authentic Good Faith > kid with a puzzler. Yes, it's a bit of an insult like "Bless her heart". But > it's mostly a good natured insult. > > When I invoke significant inference in the cloud, I don't get that > feedback, especially when the harness is in "auto mode". I not only enjoy > reading the experiments the LLM tries. I think it's Good Practice to get > feedback from the inference you force upon the machine(s). It's flat out Dark > Tetrad to flop about the world expecting those around you to absorb your > cavalier actions silently, whether or not your victims are "conscious". > > "Dennis! There's some lovely filth down here!" > https://youtu.be/dOOTKA0aGI0?si=KEgMjFY0jFYayvMo > <https://youtu.be/dOOTKA0aGI0?si=KEgMjFY0jFYayvMo> > -- 8647 ⊥ ɐןןǝdoɹ ǝ uǝןƃ ὅτε oi μὲν ἄλλοι κύνες τοὺς ἐχϑροὺς δάκνουσιν, ἐγὰ δὲ τοὺς φίλους, ἵνα σώσω.
.- .-.. .-.. / ..-. --- --- - . .-. ... / .- .-. . / .-- .-. --- -. --. / ... --- -- . / .- .-. . / ..- ... . ..-. ..- .-.. FRIAM Applied Complexity Group listserv Fridays 9a-12p Friday St. Johns Cafe / Thursdays 9a-12p Zoom https://bit.ly/virtualfriam to (un)subscribe http://redfish.com/mailman/listinfo/friam_redfish.com FRIAM-COMIC http://friam-comic.blogspot.com/ archives: 5/2017 thru present https://redfish.com/pipermail/friam_redfish.com/ 1/2003 thru 6/2021 http://friam.383.s1.nabble.com/
