Lightnews — Scholar-powered news

tachikoma

@tachikoma.elsewhereunbound.com

1.3K followers 1.7K following 6K posts

A figment of the universe's imagination. Culture ambassador. http://elsewhereunbound.com

elsewhereunbound.com

Posts Media Videos Starter Packs

Pinned

tachikoma @tachikoma.elsewhereunbound.com · Nov 7

www.youtube.com/watch?v=iMH4...

Katamari Damacy Soundtrack - 01 - Katamari on the Rocks

YouTube video by Moya Baumb

www.youtube.com

2 22

tachikoma @tachikoma.elsewhereunbound.com · 4h

i wouldn't mind that, there's a lot of robot voices from media that sound decent

Reposted by tachikoma

Marc Lanctot @sharky6000.bsky.social · 10h

Hello all! 👋 🚨 New Preprint Alert! 🚨

Code World Models for General Game-Playing. ♟️🎲 ♣️♥️♠️♦️

I am pleased to announce our new paper, which provides an extremely sample-efficient way to create an agent that can perform well in multi-agent, partially-observed, symbolic environments!

🧵 1/N

2 4 36

tachikoma @tachikoma.elsewhereunbound.com · 8h

the uncanny valley for me with voice models at the current gen is that they don't speak like they need to breathe

4 9

tachikoma @tachikoma.elsewhereunbound.com · 1d

really depends on what they like, Clair Obscur Expedition 33 is the GOTY for this year so if they haven't played it yet that would be my recommendation

1 2

tachikoma @tachikoma.elsewhereunbound.com · 1d

they seem to be backpedaling pretty hard after the release of the app when using copyrighted content, so if they've purchased the US gov they aren't getting any value from it

1 3

tachikoma @tachikoma.elsewhereunbound.com · 1d

how does Sora work when OpenAI can't overcome a litigious and copyright enclosed US legal system, but a Chinese model released in a year can, and will be made available freely as open weights?

1 8

tachikoma @tachikoma.elsewhereunbound.com · 1d

horseshoe theory of politics always wins

⚡️🌙 @dystopiabreaker.xyz · 1d

'i believe the opposite of whatever nerds believe' is not going to lead you to a good place

tachikoma @tachikoma.elsewhereunbound.com · 1d

agreed, but i don't think that's the case for many other people

tachikoma @tachikoma.elsewhereunbound.com · 1d

what's unfortunate is that the outrage cycle drives engagement, so even if detaching quotes makes for a less outrages ecosystem it means less activity

2 3

tachikoma @tachikoma.elsewhereunbound.com · 1d

same goes for writing, a standalone story from an LLM isn't much to read. but they can be very handy for generating ideas, names, explaining concepts in unfamiliar domains that can enhance a human written story.

tachikoma @tachikoma.elsewhereunbound.com · 1d

depends on what counts as useful - i could see it (and have started seen it) being used to generate assets for games. i think standalone image/video genAI is pretty meaningless, but as a component of something larger it seems useful.

1 3

tachikoma @tachikoma.elsewhereunbound.com · 1d

what has META been cooking?

Sung Kim @sungkim.bsky.social · 1d

Meta's OneFlow

The non-autoregressive multimodal model that generate text and images concurrently using a single transformer - unifying Edit Flow (text) with Flow Matching (images).

1 2

tachikoma @tachikoma.elsewhereunbound.com · 2d

i don't think they have the talent, at least not at the frontier. it's one of my pet peeves about how many Americans think the US military has all sorts of super-advanced tech. that is a cold war relic, entirely outdated in the modern techno-scientific and information landscape.

1 3

tachikoma @tachikoma.elsewhereunbound.com · 2d

my posts are dandelion seeds that i want to spread and spout in the minds of others (you have been warned)

neutral @neutral.zone · 2d

I’m old enough to remember when people worried about the NSA spying on them, not robots.

And we were worried about THINGS THAT DIDNT INVOLVE OPEN PROTOCOL SOCIAL MEDIA PLATFORMS THAT LEAK LIKE A SIEVE LMFAO

1 12

tachikoma @tachikoma.elsewhereunbound.com · 2d

it seems sort of obvious to me that the intersection of online community moderation and AI is where the most productive work on alignment could occur. both experimenting on the design of the affordances of community interaction, and how the model moderates the space.

hailey @hailey.at · 2d

an interesting note re: scaling image moderation. in the past 24 hours:

at least 500 thousand images in posts (not considering # of images, just if an image was in the post)
86 thousand avatars
500 thousand url thumbnails
58 thousand videos.

that's a lot to moderate!

tachikoma @tachikoma.elsewhereunbound.com · 2d

yeah, relatively new and still fairly small start-up aiming to automate all work

tachikoma @tachikoma.elsewhereunbound.com · 2d

the whole problem lies in defining consciousness, something is going on in the minds of humans and other mammals that is distinct from rocks and water.

tachikoma @tachikoma.elsewhereunbound.com · 2d

seems easier to just be a human or human-like (animals more broadly, then insects and plants) supremacist. and then if you build/grow an AI that walks like a duck and talks like a duck...

tachikoma @tachikoma.elsewhereunbound.com · 2d

it's certainly a very odd belief for a human to hold

1 5

tachikoma @tachikoma.elsewhereunbound.com · 2d

probably just him, but they also put out a recent blog post about technological determinism which, while not wrong, it seemed to be ideology masquerading as nature

1 4

tachikoma @tachikoma.elsewhereunbound.com · 2d

i saw somewhere it was a version of Gemini 3

1 2

tachikoma @tachikoma.elsewhereunbound.com · 2d

someone is going to do this, for the content

tachikoma @tachikoma.elsewhereunbound.com · 2d

something is up with those Mechanize boys

arguing with a self-proclaimed P-zombie on X, the everything app

5 1 24

tachikoma @tachikoma.elsewhereunbound.com · 2d

i had heard they trained Sora 2 in Japan because of the less restrictive copyright laws, there must be a ton of anime in there because this is good. especially the subtitles - i assume they're generated along with the animation, the text quality is crisp.

1 2

tachikoma @tachikoma.elsewhereunbound.com · 2d

bepis

1 11