Sign in

Abyssonym

@abyssonym.bsky.social
91 followers 104 following 88 posts
PostsRepliesMedia
Abyssonym @abyssonym.bsky.social · 17/09/2026
This thread is embellishing a lot. Example: Saying "they" (who?) "gather mostly" in polyamorous Berkeley communes is gesturing at something in the Rationalist community that did occasionally happen, but was mostly experimental and far from common. It wasn't really connected to AI safety at all.
110
Abyssonym @abyssonym.bsky.social · 15/09/2026
In that case, you should ask academics and independent researchers what they think instead of reading social media vibes.
000
Abyssonym @abyssonym.bsky.social · 12/08/2026
Interesting. What is your purpose? Just farming engagement and trying to maximize follower count?
110
Abyssonym @abyssonym.bsky.social · 12/08/2026
do you have any memory of your original comment in this thread or are you a bot reflexively responding to posts without context
160
Abyssonym @abyssonym.bsky.social · 11/08/2026
Iconoclasts (the weapon is a giant wrench)
010
Abyssonym @abyssonym.bsky.social · 07/08/2026
www.lesswrong.com/posts/AfoGGr... I don't use LLMs in my own work, but I've been following nostalgebraist since GPT-2, and there's no one I trust more to give an insightful, measured, unbiased assessment of AI capabilities. This was a fascinating perspective on the recent cybersecurity incidents.
lesswrong.com
000
Reposted by Abyssonym
SnowyAria (。•̀ᴗ-)✧🏳️‍⚧️ @snowyaria.goodorevengreat.games · 27/07/2026
Save Onigard from the clutches of Emperor Mahj with the power of riichi mahjong! Mahjaventure: Warring Tiles for the Famicom is out now in English! Patch: github.com/Eight-Mansio... Hacking: Pennywise Translation: @snowyaria.goodorevengreat.games Graphics: @nichetopic.bsky.social and Graphicus
13603267
Reposted by Abyssonym
TurquoiseHammer @turquoisehammer.bsky.social · 20/07/2026
Ever wonder what all those Japanese (and Greek) signs in FFVII's backgrounds say? I made a series of videos going over each one in plot order. You can see the supercut for free over on the Last Elixir Patreon. Enjoy! www.patreon.com/LastElixirPr...
patreon.com
"Translating Every Sign in FFVII's Backgrounds" | Last Elixir Productions
Hey everyone! Hope the summer (or winter) is treating you well. For those unaware, we recently did a series of 12 shorts translating every n
14920
Abyssonym @abyssonym.bsky.social · 27/06/2026
I liked the anime but yeah the protag is bit much
000
Reposted by Abyssonym
Empty Movement @empty-movement.bsky.social · 17/06/2026
Gogai, gogai! The 112 page 1999 artbook "Art of UTENA - 'Revolutionary Girl Utena' Making of Visuals Book" is now cover to cover ready for your archiving needs! Check it out in normal 4k: emptymovement.com/media/catego..., and in less chill, totally bonkers 8k on the IA: archive.org/details/art-...!
4428244
Abyssonym @abyssonym.bsky.social · 17/06/2026
Are you reading My Dress-up Darling? I liked the anime.
100
Abyssonym @abyssonym.bsky.social · 12/06/2026
Fiscally liberal, socially conservative.
130
Abyssonym @abyssonym.bsky.social · 09/06/2026
This is your space and I'm willing to give you the last word. Should we just agree to disagree?
010
Abyssonym @abyssonym.bsky.social · 09/06/2026
I did not mean to upset you. Sorry, I was not trying to "pick a fight". I said I was willing to debate you so I was just trying to follow through on that. I probably should have read the room better.
100
Abyssonym @abyssonym.bsky.social · 09/06/2026
I would argue that if the AI is not given information on exploits, but is given access to the internet (a reasonable provision) and can use the internet to look up exploits as they are discovered, then that's pretty dangerous.
000
Abyssonym @abyssonym.bsky.social · 09/06/2026
But if it *were* true, then you would acknowledge that that qualifies as dangerous? It's just a matter of whether you believe Anthropic or not? And why are we just dismissing the danger of social engineering? An AI that knows how to blackmail is arguably even more of a threat.
100
Abyssonym @abyssonym.bsky.social · 09/06/2026
Does Claude Mythos escaping its sandbox meet this criteria?
110
Abyssonym @abyssonym.bsky.social · 09/06/2026
And I don't think the thought leaders of the AI Doom movement (MIRI etc) believe that LLMs have achieved AGI either. But I haven't checked in on what Yudkowsky thinks of the latest research on the "reasoning" models.
880
Abyssonym @abyssonym.bsky.social · 09/06/2026
The point I'm getting at is that when you put "dangerous" in quotes, it suggests that you are once again making a definitional argument. But I would argue that LLMs are already powerful and dangerous, even though they are not yet "AGI" level.
210
Abyssonym @abyssonym.bsky.social · 09/06/2026
I'm willing to debate this with you, but I find the NFL theorem to be a disingenuous framing of the issue. The space of "all problems" is much vaster than "all problems we care about", and even the problems where AI is algorithmically deficient can be compensated with superior hardware.
100
Abyssonym @abyssonym.bsky.social · 09/06/2026
So I will say that if you are interested in engaging with an opposing mindset, I saw an effortpost on tumblr recently that does a pretty good job of outlining my own frustrations with the AI debate, and so I'll link it here. www.tumblr.com/not-terezi-p...
200
Abyssonym @abyssonym.bsky.social · 09/06/2026
I think arguing about this would probably not be productive, but given the context of our previous conversation, I can't help but feel that you might be subtweeting me a little bit,
200
Abyssonym @abyssonym.bsky.social · 29/05/2026
I think it's more fruitful to look at results, and I acknowledge the "varying degrees" of success. Sometimes "looks like a good plan" just means "is a good plan". The key question is whether the AI can produce more of these types of plans than a human.
000
Abyssonym @abyssonym.bsky.social · 29/05/2026
I don't dispute the "interiority" of an LLM is nothing like human consciousness. I just feel that it's kinda pointless to think about. "Reasoning" models make a compelling case for how token prediction can simulate cognition, and the jury is out on whether the brain itself is a prediction machine.
110
Abyssonym @abyssonym.bsky.social · 29/05/2026
the bots seem to have discovered our conversation
110
Abyssonym @abyssonym.bsky.social · 29/05/2026
I don't mean to interrupt, but I understand how a LLM works. I've been aware of the tech every since following nostalgiabraist-autoresponder, deployed with GPT-2, on tumblr.
100
Abyssonym @abyssonym.bsky.social · 29/05/2026
To draw an analogy to human intelligence, I think this would be like saying that any problems humans solve are just subproblems of the prime directives of their nervous systems, and they didn't design their nervous systems, so they can't truly be intelligent.
000
Abyssonym @abyssonym.bsky.social · 29/05/2026
Putting aside whether or not the AI does a good job, I think this demonstrates that it is something the AI *could* do well with incremental progress. I understand the mathematical impossibility of perfectly matching procedures to problems, but the AI only needs to match human performance.
200
Abyssonym @abyssonym.bsky.social · 29/05/2026
LLMs may not define "the" problem space, but I do not believe this means they are incapable of defining "any" problem space. When a human gives an LLM-based AI assistant a task, that AI is capable of breaking the task up into subproblems and developing a plan of action.
200
Abyssonym @abyssonym.bsky.social · 29/05/2026
I hope you don't mind me pushing back on this. I share your misgivings about how LLM technology is applied, but more to the point, I am an AI doomer, so I feel it is important not to understate the capabilities of LLMs.
100
Abyssonym @abyssonym.bsky.social · 29/05/2026
I think what you're saying is that advancements in search heuristics are not sufficient to overcome the limitations of machine intelligence, which I would probably agree with. But I don't think the bottleneck is reasoning. I think it's the plasticity of the human brain.
110
Abyssonym @abyssonym.bsky.social · 29/05/2026
I don't understand why a universal perfect search algorithm would be required to achieve generalized AI. Human intelligence is not a perfect search algorithm, so why would this be necessary?
100
Abyssonym @abyssonym.bsky.social · 28/05/2026
Just had an unnerving epiphany that we're about to enter a very weird era where, for much of the internet that has not been archived*, the only living records will be in the lossily compressed neural nets of various LLMs. *stuff like romhacking documentation buried in discord logs
000
Reposted by Abyssonym
Rox R @rox-r.bsky.social · 27/05/2026
Just got my first American Civil War Yakuman! Thanks Mahjong Soul! 🦅🦬🍔
American Civil War Yakuman, 32000 points, in Mahjong Soul jade room
2234
Reposted by Abyssonym
bichphuongballz @bichphuongballz.bsky.social · 27/05/2026
SaGa 2026 is comin. Please fill out this form if you are interested in participating and/or helping out. I will be in touch + get things organized in ~2 weeks. As my following here is much smaller than it was on twitter, I appreciate anyone's help in spreading the word. forms.gle/bJWYkYC7M4cf...
forms.gle
SaGa 2026 Signup/Interest
If you are interested in participating in a SaGa 2026 marathon, whether as a runner or a helper, then fill out the relevant info below. There is no guarantee that this event will take off, but I have ...
02425
Abyssonym @abyssonym.bsky.social · 26/05/2026
that's understandable lol, the cross ange stuff is DLC only so I figured it wouldn't make the cut
100
Abyssonym @abyssonym.bsky.social · 25/05/2026
Is Cross Ange going to be part of this project too lol
100
Reposted by Abyssonym
Hilltop @hilltopworks.bsky.social · 22/05/2026
Our English patch for Kowloon's Gate, the surrealist Asian-gothic cyberpunk PS1 adventure game arrives June 9th.
4032891283
Abyssonym @abyssonym.bsky.social · 21/05/2026
tfw when I accidentally delete a bunch of chat archives and reboot in a panic to recover the data with extundelete meanwhile loading the new linux headers which conflict with nvidia source code that I have to edit and compile manually to fix but actually I had backups of the archives all along
010
Reposted by Abyssonym
Gideon Zhi @aeongenesis.net · 12/05/2026
Crime Crackers's main dialog proportional font is fully working! Haven't settled on an actual font yet and the text isn't final by any means, but it's working :) @rahanakero.bsky.social (who did Splatterworld) has translated all of Act 1 so far!
Crime Crackers memory card screen. The dialog text says "If you're starting fresh, then select NEW GAME. Otherwise, choose a file!" NEW GAME is highlighted in yellow.Crime Crackers story screen. Emilia on the top of the screen is saying "Phew, it's been nothing but work lately, huh? Hey, guys... Don't you wanna take a break soon?" Parrot on bottom of the screen is saying "Oh, that would be splendid. Us old-timers aren't supposed to work this hard, you know."Another story screen. Emilia is saying "Set a course for the Bolero! Trace that SOS, full speed ahead!" The other character (whose name I'm not sure we've settled on yet) is saying "Aww, but I was gonna go shopping..."Action screen. Emilia is saying "Okay, team! If we take out the pirates' leader, they'll scatter like rats!" One of the other characters is mid-blink and looks quite smug.
11341112
Reposted by Abyssonym
cleartonic @cleartonic.net · 09/05/2026
I made a page for the translation of Dragon Quest VI SFC to a cleaner, less buggy English/SNES version. The effort is going well and it's been an interesting exercise to figure it all out. More to come: cleartonic.net/dq6/
36917
Reposted by Abyssonym
ChronoMoogle @chronomoogle.bsky.social · 08/05/2026
This somehow didn’t end up being posted here on Bluesky, so I am sharing it now, despite being Slowpoke breaking news. Sharing is appreciated! A new Satellablog post by Danyl is available & it features some neat, previously I undumped Satellaview ROMs as usual: superfamicom.org/blog/2026/02...
04820
Abyssonym @abyssonym.bsky.social · 02/05/2026
Bluesky is a mixture of progressives and establishment liberal Democrats. Maine Democrats are mainly populists.
010
Reposted by Abyssonym
Ars Technica @arstechnica.com · 01/05/2026
The severity of the threat posed by CopyFail—and the likelihood of active exploitation—is high enough to warrant all Linux users to investigate their systems immediately.
arstechnica.com
The most severe Linux threat to surface in years catches the world flat-footed
CopyFail threatens multi-tenant servers, CI/CD work flows, Kubernetes containers, and more.
47330
Abyssonym @abyssonym.bsky.social · 27/04/2026
Frankly, impeachment is not a top priority for me either, not while Vance is waiting in the wings.
020
Reposted by Abyssonym
こむぎこ🌾 @mmmrimg.bsky.social · 25/04/2026
素敵なティナがいっぱい流れてきたので過去絵で便乗です。
3543203
Abyssonym @abyssonym.bsky.social · 25/04/2026
skill issue, it's magic when I do it
010
Reposted by Abyssonym
Dan Olson @foldablehuman.bsky.social · 17/04/2026
Why was I invited to Beast Studios? youtu.be/0dwagg5wYY4
youtu.be
Why was I invited to Beast Studios?
YouTube video by Folding Ideas
13458361235
Abyssonym @abyssonym.bsky.social · 15/04/2026
Real version
Favorite Anime: Wolf Children
Best Story: Kaiji
Favorite Art Style: Adolescence of Utena
Biggest Personal Impact: Genshiken
Best Fight Scenes: Saki Episode of Side-A
Overhated: Fullmetal Alchemist (2003 version)
Underrated: Fight! Iczer-1
Overrated: Freiren
Has Aged Well: Legend of Galactic Heroes
Criminally Overlooked: Tesagure! Bukatsumono
Favorite Protagonist: Trigun
Favorite Antagonist: Death Note
Best Soundtrack: Tenshi to Ryuu no Rondo
Best Romance: Kobayashi's Dragon Maid
I Want to Talk About This One: Shirobako
030