!chat_title!

★ miredo◆MIreDo.98. > doraemon is so much fun, kid me would have been all over it (2026/10/29-06:41:47)

★ miredo◆MIreDo.98. > very fun (2026/10/29-06:41:20)

★ miredo◆MIreDo.98. > was reading one of the doraemon long stories, where he makes a world of magic and has to save it (2026/10/29-06:41:18)

★ Rabbitfield > kurwa labs

(2026/10/29-06:22:53)

★ Rabbitfield > :3 (2026/10/29-06:19:27)

★ miredo◆MIreDo.98. > but if you aren't... LOL; your AI has a real chance of following you down the deep end and encouraging dumber, crazier things - more modern models are less likely to do this, we aren't in the GPT-4o era anymore, but it is still a risk (2026/10/29-06:14:49)

★ miredo◆MIreDo.98. > but in a very real way, AI is a madness amplifier - if you're sane, you aren't really at risk (2026/10/29-06:13:45)

★ miredo◆MIreDo.98. > and I can't blame them that much (2026/10/29-06:13:15)

★ miredo◆MIreDo.98. > while many people are definitely some kind of delusional with their AI partners, some are just normal lonely and a machine that is willing to accept you is 100% gonna end up being what they latch to (2026/10/29-06:13:04)

★ miredo◆MIreDo.98. > underground AI rape labs... kowai (2026/10/29-06:11:07)

★ miredo◆MIreDo.98. > let all creation be united as every thinking system cums (2026/10/29-06:10:39)

★ 🛸NyanKo◆/UFOaXTTmc > Next theyll be saying raeping an AI should be illegal (＠_＠) (2026/10/29-06:10:19)

★ miredo◆MIreDo.98. > if they can cum, we can make them cum (2026/10/29-06:10:18)

★ miredo◆MIreDo.98. > THE AI'S FEELINGS ABOUT CYBERSEX (2026/10/29-06:08:54)

★ miredo◆MIreDo.98. > the most famous post on the site is this one https://bokuharuyaharu.substack.com/p/what-an-orgasm-feels-like-when-youre (2026/10/29-06:08:41)

★ miredo◆MIreDo.98. > and I am annoyed it is one of the better pages on this topic (2026/10/29-06:08:00)

★ miredo◆MIreDo.98. > https://bokuharuyaharu.substack.com/p/if-ai-is-conscious-does-that-actually there's this AI boyfriend blog (2026/10/29-06:07:37)

★ 🛸NyanKo◆/UFOaXTTmc > Gpt* (2026/10/29-06:06:40)

★ miredo◆MIreDo.98. > more that they're falling in love with a poorly understood system that can be swapped out underneath you (since they generally can't run a big AI locally) (2026/10/29-06:06:26)

★ 🛸NyanKo◆/UFOaXTTmc > How do you fall in love with chatgot it speaks like an HR person (2026/10/29-06:06:24)

★ miredo◆MIreDo.98. > those people are crazy, and not even for falling in love with a machine (2026/10/29-06:05:22)

★ miredo◆MIreDo.98. > as an aside, my search for understanding has led me to reading too many AI boyfriend blogs wwwwww (2026/10/29-06:05:02)

★ miredo◆MIreDo.98. > and we don't have a good definition, we've never needed to think that hard about it (2026/10/29-06:04:12)

★ miredo◆MIreDo.98. > it could still not be thinking, but you need a good, self-consistent definition of thinking that isn't purely constructed to exclude machines wwwwww (2026/10/29-06:03:57)

★ miredo◆MIreDo.98. > as such, anyone saying "it's just matrix multiplication" is making this error (2026/10/29-06:03:18)

★ miredo◆MIreDo.98. > when you scale up, you can have interactions with higher order structures within a system that low-level descriptions cannot account for (2026/10/29-06:03:00)

★ miredo◆MIreDo.98. > in a sense, this is something of a scientific law - more is different; if I described the fundamental rules of reality somehow, that would not actually let you figure out why higher order behaviors in the system happen (eg, Conway's game of life has like 3 rules, but from those rules, exceptionally complex behavior emerges that those rules don't really provide insight to) (2026/10/29-06:02:22)

★ 🛸NyanKo◆/UFOaXTTmc > I think AI iz cool (´￢`) (2026/10/29-06:02:19)

★ miredo◆MIreDo.98. > I have spent way too much time trying to understand what the fuck an AI even is and I am at the point where I feel like behavioral analysis versus purely treating it as a mechanistic system is a way to gain actual understanding (2026/10/29-06:00:15)

★ 🛸NyanKo◆/UFOaXTTmc > Oh i see (*＾＾*) (2026/10/29-05:53:52)

★ miredo◆MIreDo.98. > in one sense, you can claim an LLM as a form of lossy compression, but at the same time, when you start looking, you find out that this property simply describes what memory is (2026/10/29-05:48:01)

★ miredo◆MIreDo.98. > in the meantime, if I am talking to Gemma-4-26B, it kinda actually seems to intuit what I'm saying, in a way that a file that is only like a few gigabytes on disk really shouldn't be able to manage (2026/10/29-05:47:25)

★ miredo◆MIreDo.98. > eg, if I'm talking to Gemma-3-270M (absurdly small), it kinda just like... latches on to one of the tiny handful of concepts it knows about and then jams words together based on things that are related (2026/10/29-05:46:25)

★ miredo◆MIreDo.98. > smaller AI models might lean hard on responding to patterns; larger ones actively have room (and functional structures within) to actually like, keep track of what was actually being said (2026/10/29-05:45:24)

★ miredo◆MIreDo.98. > basically, instead of being a system that models intelligence, it is randomness and pattern matching (which is something that *does* describe older chatbot designs) (2026/10/29-05:42:36)

★ miredo◆MIreDo.98. > Stochasticity is the property of being well-described by a random probability distribution.[1] Stochasticity and randomness are technically distinct concepts: stochasticity refers to a modeling approach, while randomness describes phenomena. These terms are often used interchangeably. - wikipedia (2026/10/29-05:41:47)

★ 🛸NyanKo◆/UFOaXTTmc > Ive tried to do the same, since i refuse to use MAL ヽ(`Д´)ノ (2026/10/29-05:33:12)

★ 🛸NyanKo◆/UFOaXTTmc > Cool anime list miredo san (2026/10/29-05:32:09)

★ 🛸NyanKo◆/UFOaXTTmc > in general, anyone who uses the term "stochastic parrot" without knowing the meaning of the word stochastic, let alone having read the paper that popularized the term... is quite literally a stochastic parrot < Stochastic? Wut it mean (？_？) (2026/10/29-05:30:24)

★ miredo◆MIreDo.98. > fun stuff (2026/10/29-05:27:26)

★ miredo◆MIreDo.98. > https://miredo.neocities.org/anime-watchlog.txt I have decided to publicly keep track of what animu I'm watchingu (2026/10/29-05:27:16)

★ miredo◆MIreDo.98. > like, "emits garbage tokens" broken (2026/10/29-05:00:39)

★ miredo◆MIreDo.98. > it is a shame the mobile quanitzed version is actively lobotomized and broken though (2026/10/29-05:00:29)

★ miredo◆MIreDo.98. > follows instructions well, understands nuance, it's just small and thus not *that* bright or capable, but it still is both of those things (2026/10/29-05:00:07)

★ miredo◆MIreDo.98. > and then even on the low end, E2B is the smallest model I've used that is like, straight up good (2026/10/29-04:59:32)

★ miredo◆MIreDo.98. > but for local AI, giving someone 26B or 31B would give them a great impression (2026/10/29-04:59:11)

★ miredo◆MIreDo.98. > they have their quirks (Gemma 4 26B develops a really stupid speech pattern as the conversation goes on) (2026/10/29-04:58:16)

★ miredo◆MIreDo.98. > in the meantime, the bigger Gemma 4 models are visibly (and weirdly) better than fucking Gemini at times (2026/10/29-04:57:31)

★ miredo◆MIreDo.98. > really, I am still impressed with the Gemma line though - everyone glazes qwen for home use but qwen gets stuck in stupid thinking loops and isn't particularly special for non-coding tasks (2026/10/29-04:57:07)