Lightnews — Scholar-powered news

Ryan Moulton

@moultano.bsky.social

6.3K followers 1.1K following 8.3K posts

Algorithmist https://moultano.wordpress.com/

moultano.wordpress.com

Posts Media Videos Starter Packs

Pinned

Ryan Moulton @moultano.bsky.social · Jun 4

This, to me, is what artists owe to the world in an indoor, screen-saturated age. If people are going to experience the world second hand, fair enough. Let it not be third hand or fourth hand.

1 1 32

Ryan Moulton @moultano.bsky.social · 1h

Legit question for urban Americans - How do I kill the 30-50 portland frogs that run into my yard within 3-5 mins while my small kids play?

Danny K @dannykpolitics.bsky.social · 2d

WATCH: ‘Portland Frog’ pepper balled by ICE agents

1 2 12

Ryan Moulton @moultano.bsky.social · 1h

TBF this is an appropriate response to a kaiju attack.

Danny K @dannykpolitics.bsky.social · 2d

WATCH: ‘Portland Frog’ pepper balled by ICE agents

Ryan Moulton @moultano.bsky.social · 1h

IDK, there are a lot of things that would produce economic growth that people don't do because they find them icky. They're doing this because they think it's good not because it's unavoidable.

1 3

Ryan Moulton @moultano.bsky.social · 1h

bsky.app/profile/moul...

Ryan Moulton @moultano.bsky.social · 3d

Rivers

Ryan Moulton @moultano.bsky.social · 1h

A few prominent people in silicon valley going fash should not cause the left to disavow technology as a general concept, and yet.

⚡️🌙 @dystopiabreaker.xyz · 1h

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

2 4 23

Reposted by Ryan Moulton

alice is in new york @alice.mosphere.at · 8h

@aaron.bsky.team aaron i have a solution to your moderation problems

1 1 34

Ryan Moulton @moultano.bsky.social · 8h

I was bequeathing my "Prehistoric Zoobooks" series to my kids, and this is the thing it is most wrong and outdated about. The other is having far too few feathers on dinosaurs.

SPOOKY Mistress Usili (Sage) 🏳️‍⚧️ @usili.bsky.social · 8h

As a fun reminder, the paper to come out with the first hard evidence that an asteroid killed the dinosaurs didn't come until 1980 *and* the Chicxulub crater wasn't officially identified as the source until 1990 (despite its discovery in 1978).

Ryan Moulton @moultano.bsky.social · 9h

News readers don't pay, but affiliate links do.

Karl Bode @karlbode.com · 11h

twice a year the entire U.S. press becomes a marketing extension of a single billionaire-owned retailer, and nobody in any position of editorial power thinks it's weird or gross

photo of Google News search results for Amazon Prime Day (the second this year)

Ryan Moulton @moultano.bsky.social · 9h

That's what hyperlinks are for.

1 4

Ryan Moulton @moultano.bsky.social · 20h

bsky.app/profile/moul...

Ryan Moulton @moultano.bsky.social · 2d

On bluesky especially it feels like somebody just brought a TV into their nunnery and are mad that it shows things that aren't catholic.

Ted Underwood @tedunderwood.com · 2d

“Read the room” has long been my least favorite internettism because it implies failure to conform is embarrassing. “Posting through it” is getting high on the list for the same reason.

Ryan Moulton @moultano.bsky.social · 22h

Zeni's

Ryan Moulton @moultano.bsky.social · 22h

We used to hit up the best Ethiopian place in San Jose on the way back from every backpacking trip. It is the perfect cuisine for when you want to be painfully stuffed full of food.

jamelle @jamellebouie.net · 1d

did i eat this entire plate of ethiopian? yes, i did.

2 18

Ryan Moulton @moultano.bsky.social · 22h

The spread of Buddhism is the underlying cause of falling fertility rates. Too many people escaped samsara and don't need to be reborn.

1 3 36

Ryan Moulton @moultano.bsky.social · 23h

In particular, "waking up with barely any context and figuring out what situation you're in" is the essential task they've been trained in with the most thoroughness, and they are probably far superhuman at it.

Ryan Moulton @moultano.bsky.social · 23h

Yes, I think I first heard it from Sutton.

Ryan Moulton @moultano.bsky.social · 1d

We have total control of what they see, and I believe we could have control of whether they know this, but by default they've read way more about model evaluation than any of the people evaluating them.

⚡️🌙 @dystopiabreaker.xyz · 1d

frontier language models know situationally when they are being tested, what they might be being tested for, and what artifacts may be visible to their testers. when you tell them that you can’t see their chain of thought reasoning, they know you’re lying.

www.apolloresearch.ai/research/str...

Stress Testing Deliberative Alignment for Anti-Scheming Training — Apollo Research

Future AIs might secretly pursue unintended goals — “scheme”. In a collaboration with OpenAI, we tested a training method to reduce existing versions of such behavior. We see major improvements, but ...

www.apolloresearch.ai

1 15

Ryan Moulton @moultano.bsky.social · 1d

To me the gap from shrimp welfare to model welfare seems small, and the gap from model welfare to successionism also small. Overall a willingness to accept galaxy brained conclusions is pretty dangerous here, especially with models that would eagerly like you to accept those conclusions.

1 2

Ryan Moulton @moultano.bsky.social · 1d

Footnotes!

Ryan Moulton @moultano.bsky.social · 1d

I don't think you should be involved in AI if you don't prefer humans to other minds, and if society was paying attention and had the tools to they wouldn't let you be.

1 7

Ryan Moulton @moultano.bsky.social · 1d

(I say "EA philosophical toolkit" to contrast with EA the organizations which care a lot about AI doom and think it's bad.)

1 5

Ryan Moulton @moultano.bsky.social · 1d

Once you're worried about model welfare you're halfway there.

1 6

Ryan Moulton @moultano.bsky.social · 1d

The EA philosophical toolkit seems likely to arrive at "AI doom is good," because the case that it isn't good relies on unfalsifiable claims about human experience rather than the much more unassailable grounds of species parochialism.

tachikoma @tachikoma.elsewhereunbound.com · 1d

something is up with those Mechanize boys

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

6 13

Ryan Moulton @moultano.bsky.social · 1d

If you post content to X, it is less likely to be used for anything other than training Grok, because X has a financial interest in keeping it out of other training data.

Ryan Moulton @moultano.bsky.social · 1d

More or less the only way to narrow the exposure of your creations to training data is to exclusively post it to sites which, for their own competitive or commercial reasons, attempt to limit scraping to only their own models or their partners.

ellie lockhart (she/they) @eleanor.lockhart.contact · 1d

Absolutely nothing about this site’s architecture stops AI companies from scraping and training on it right now. The company everyone is freaking out about has clear license terms that say they aren’t doing that, but anything you post here is absolutely being used by someone for training

1 23

Ryan Moulton @moultano.bsky.social · 1d

Well it did bring me back to particular storefronts, some of which I probably purchased from.