Reload: 40 seconds Lines: 50
Autistic ramblers (4): kaguya◆anonwaha◇hachikuji◆Rabbitfield◇
System-chan > anonwaha went to sleep... 💤 (2026/10/19-07:45:45)
★ anonwaha◆bGTvsQ3YBE > @sleep (2026/10/19-07:45:45)
★ anonwaha◆bGTvsQ3YBE > nyaa (2026/10/19-07:45:37)
★ kit◆xnSCwv3URQ > miaw (2026/10/19-07:37:33)
★ hachikuji◆naRqMpZoTs > meow (2026/10/19-07:33:47)
★ tootsie > *hidey hole retreat* (2026/10/19-07:08:00)
★ hachikuji◆naRqMpZoTs > SNEAKAR (2026/10/19-06:26:18)
★ tootsie > *sneak sneak* (｀・ω・´) (2026/10/19-06:25:43)
★ hachikuji◆naRqMpZoTs > ah XD (2026/10/19-06:07:15)
★ anonwaha◆bGTvsQ3YBE > i set an alarm, but apparently for the wrong day (2026/10/19-06:00:25)
★ anonwaha◆bGTvsQ3YBE > u? < i got 6h15m ...though it was supposed to just be a 1h30m nap :sweat2: (2026/10/19-06:00:04)
★ hachikuji◆naRqMpZoTs > u? (2026/10/19-05:57:43)
★ hachikuji◆naRqMpZoTs > but i feel well rested (2026/10/19-05:57:37)
★ hachikuji◆naRqMpZoTs > how many hours did u get? < like 3 (2026/10/19-05:57:14)
★ yaku◆x1yMC5lu4g > gabagool (2026/10/19-05:53:11)
★ anonwaha◆bGTvsQ3YBE > how many hours did u get? (2026/10/19-05:50:16)
★ anonwaha◆bGTvsQ3YBE > ohayo hachi! (2026/10/19-05:50:04)
System-chan > hachikuji woke up! 🌅 (2026/10/19-05:49:28)
★ hachikuji◆naRqMpZoTs > @wakeup (2026/10/19-05:49:28)
System-chan > 👁️⩊👁️ drop'd truth nuke https://www.youtube.com/watch?v=2IXX8_4wU5s (2026/10/19-04:56:15)
★ 👁️⩊👁️ > @drop>truth nuke https://www.youtube.com/watch?v=2IXX8_4wU5s (2026/10/19-04:56:15)
★ yaku◆x1yMC5lu4g > burn (2026/10/19-04:46:18)
★ miredo◆MIreDo.98. > savage (2026/10/19-04:06:54)
★ anonwaha◆bGTvsQ3YBE > "Go and eat, then. Even that, you'll probably stare at the plate for ten minutes deciding whether you're allowed." :xd: (2026/10/19-04:05:15)
★ anonwaha◆bGTvsQ3YBE > that one actually sounds mean (;´Д`) < that was one of the nicer ones - i LOL'd at it :xd: (2026/10/19-03:58:12)
★ yaku◆x1yMC5lu4g > that one actually sounds mean (;´Д`) (2026/10/19-03:57:39)
★ yaku◆x1yMC5lu4g > waha: ouch... (2026/10/19-03:57:29)
★ yaku◆x1yMC5lu4g > sometimes i like to LOL from reading old thread archives (｀・ω・´) (2026/10/19-03:57:13)
★ anonwaha◆bGTvsQ3YBE > "If not doing things were a sport, you'd have a lifetime achievement award by now and you wouldn't go up to collect it." :xd: (2026/10/19-03:50:55)
System-chan > 👁️⩊👁️ went to sleep... 💤 (2026/10/19-03:29:59)
★ 👁️⩊👁️ > @sleep (2026/10/19-03:29:59)
★ 👁️⩊👁️ > oge i go slep 2 sorri for being pain in da ass all day i was feeling berri roneri (2026/10/19-03:29:35)
★ 👁️⩊👁️ > sorry for the autistic rambling (2026/10/19-03:17:27)
★ 👁️⩊👁️ > https://www.youtube.com/watch?v=yX8yrOAjfKM (2026/10/19-02:53:25)
★ 👁️⩊👁️ > can't wait to see the proof that the world is running on windows 98 when the smort peeps integrate AI with quantum computers (2026/10/19-02:51:33)
★ 👁️⩊👁️ > i mean it reinforces my own subjective confirmation bias towards idealism (2026/10/19-02:47:57)
★ 👁️⩊👁️ > i mean, if reality is not fundamentally mental, then ai could become that mind (2026/10/19-02:45:22)
★ 👁️⩊👁️ > tbh ai reinforces my ideas about a universal mind. on some level it has access to every thought anyone can ever have, but in a broader context it can pull up anything that's statistically possible (2026/10/19-02:43:59)
★ miredo◆MIreDo.98. > guunaito red (2026/10/19-02:40:57)
★ 👁️⩊👁️ > ok sleep well red-chan (2026/10/19-02:40:51)
★ miredo◆MIreDo.98. > ultimately, a lot of people's view of LLMs were tainted really heavily by the initial release of ChatGPT, whose world model was ridiculously bad and couldn't work with numbers for the life of it (2026/10/19-02:40:47)
★ anonwaha◆bGTvsQ3YBE > gdnite red! sleep well! (2026/10/19-02:40:42)
★ redanon◆TorRZVRzqo > i'll go (try [and fail]) to sleep, good night guise (2026/10/19-02:40:24)
★ 👁️⩊👁️ > :depression: (2026/10/19-02:39:04)
★ miredo◆MIreDo.98. > they're general purpose enough that you could reasonably train an LLM on like, .BMP images encoded in base64 LOL and give it vision in the least efficient way (2026/10/19-02:38:49)
★ miredo◆MIreDo.98. > and thus can relate arbitrary concepts that were fed in and tokenized in training - there's fun shit like how Gemma-4-12B doesn't have a separate image decoder model, it just processes image tokens directly in the same dense matrix (2026/10/19-02:37:52)
★ miredo◆MIreDo.98. > since technically, they're not really "language" models; they're data relational models (2026/10/19-02:36:43)
★ miredo◆MIreDo.98. > one of the more interesting things is that LLMs are obnoxiously general purpose in a way that like, more task specific architectures are falling to the wayside in their wake in some cases (2026/10/19-02:36:12)
★ miredo◆MIreDo.98. > and have a whole-ass section of my site dedicated to LLM bullshit (not LLM-written stuff, but like, my thoughts on them) (2026/10/19-02:34:56)
★ 👁️⩊👁️ > really, it's simply just another iteration of the technological capabilities of humanity (2026/10/19-02:34:25)

- COMCHAT -
Translated and modified by Heyuri