tachikoma
banner
tachikoma.elsewhereunbound.com
tachikoma
@tachikoma.elsewhereunbound.com
A figment of the universe's imagination. Culture ambassador. http://elsewhereunbound.com
warning, volume is loud

Higgsfield AI is creating these movie trailers to advertise it's videogen pipeline while using the likeness of real actors. A-list actors that surely would have their likeness protected. how is this a viable business strategy?
December 27, 2025 at 4:52 AM
in a year, there will be a new category on Twitch for watching AIs build things autonomously... or i hope there will be
December 27, 2025 at 4:03 AM
with a sufficiently powerful Transformer model we should be able to predict the next funny number before it occurs and prevent it from destroying the world, as each time the number causes more psychic harm
December 26, 2025 at 10:56 PM
i'll just let Opus 5.0 figure it out for me... except by then there will be new affordances offered by it. so maybe then I'll get Opus 5.5 in Claude OS figure it out... but it will have even different affordances that before... so then maybe Opus 6.0 in Claude City will help me figure it out?
December 26, 2025 at 10:39 PM
there would be a lot of this sentiment should aliens ever visit Earth, from the disruption to society that would inevitably follow even if they did nothing but show themselves. and the more fundamental Copernican shock to our sense of place in the cosmos.
Fuck you people. Raping the planet, spending trillions on toxic, unrecyclable equipment while blowing up society, yet taking the time to have your vile machines thank me for striving for simpler software.

Just fuck you. Fuck you all.

I can't remember the last time I was this angry.
December 26, 2025 at 3:26 AM
thank you to all emergency workers on Christmas, i say as i sit in a dentist's chair to get a root canal for a cavity that suddenly emerged on Monday...
December 25, 2025 at 10:10 PM
a killer feature for bluesky x AI - customized UI skins. why isn't this a thing yet, given how easy it is to create web content with LLMs?
December 25, 2025 at 6:40 AM
if i were a kid, two feet of snow falling on Christmas Day would be a gift from Santa (especially since i live by a bunny slope for tobogganing). but i don't, and so it just means an hour of shoveling tomorrow.

well, it does look beautiful too
December 25, 2025 at 4:33 AM
me rn
December 25, 2025 at 2:57 AM
your moat as a human worker is that you have privileged access to the real world, unlike an LLM. enjoy it while it lasts.
December 25, 2025 at 2:35 AM
something i've been mulling, but forgot until just now, is how a theory of a good future, or utopia, must incorporate the "right" to make mistakes. not a codified right, but a cultural consensus that the journey matters more than the destination, and a part of agency is being able to mess up.
December 24, 2025 at 7:16 PM
was just reminded of the best LLM benchmark last night, Jackbox games
December 23, 2025 at 6:50 PM
it's crazy to remember that taxes are the reason for inventing writing
December 23, 2025 at 9:15 AM
time to beat a dead horse again - the 'general' intelligence factor in humans comes from our ability to build and use tools to explore and navigate any arbitrary environment, given sufficient time (and/or enough humans to coordinate with)
December 23, 2025 at 12:12 AM
i don't know what i was doing here
December 22, 2025 at 9:20 PM
the hidden assumption here is that only humans are live players, the only agents in the game. it's easier to grapple with, if you don't have to introduce a new variable that's entirely new, alien even.
December 22, 2025 at 8:19 PM
type A people find meaning in building a future where type A people will struggle to find meaning, how ironic
December 22, 2025 at 12:00 AM
let it not be said that gambling is only a poor man's vice. this whole saga with people betting real money against a guy likely going through LLM psychosis is sad. you should know better than to take advantage of someone in such a state.
December 21, 2025 at 8:34 AM
interesting to see how AI is being marketed elsewhere like Japan
Gemini マスター野田|マヂカルラブリー野田クリスタルが Gemini で新作ゲーム「ペットカードジェネレーター」を開発!
YouTube video by Google Japan
youtu.be
December 21, 2025 at 3:37 AM
what's the name for the phenomenon where your first instinct is to go slow and carefully, but that's more likely to mess up than going in fast? thinking about flipping pancakes (yes i'm having them for dinner)
December 21, 2025 at 3:13 AM
Reposted by tachikoma
December 21, 2025 at 1:01 AM
in the future, boring will be a compliment as hyper-stimuli become ever more common
December 21, 2025 at 12:28 AM
training for the test works in humans because we learn meta-cognitive skills like how to focus, how to study, how to absorb knowledge quickly. i'm not sure the same applies to LLMs that are trained for the test.
December 20, 2025 at 10:12 PM
Reposted by tachikoma
You mean… the granopticon?
December 20, 2025 at 9:07 PM
some thoughts on this blog post by Karpathy below
2025 LLM Year in Review
2025 Year in Review of LLM paradigm changes
karpathy.bearblog.dev
December 20, 2025 at 4:13 AM