Lightnews — Scholar-powered news

Reposted by Danyel Fisher

GHOULLIKEHELLMACHINE @golikehellmachine.com · 10d

more and more frequently, trump makes comments that make it seem like he is vaguely aware of the fact that his entire inner circle and cabinet are all scheming viziers, he's just sort of indifferent about it

Trump referenced a weekend conversation with Oregon Gov. Tina Kotek, and he alluded to being told by Kotek that the reality in Portland is different from what's being portrayed to him.

"I spoke to the governor, she was very nice," Trump said. "But I said, 'Well wait a minute, am I watching things on television that are different from what's happening? My people tell me different.' They are literally attacking and there are fires all over the place...it looks like terrible."

31 160 1.1K

Danyel Fisher @danyelf.bsky.social · 13d

Maybe it’s labeling things that’s the real problem. Let your data revel in their anonymity.

1 1

Danyel Fisher @danyelf.bsky.social · 13d

Cool! Then you can implement postscript over it! (I spent several years writing graph drawing algorithms, and label placement terrifies me.)

1 1

Danyel Fisher @danyelf.bsky.social · 23d

Sure: it can be useful (as here!) to show two temporally aligned phenomena. You need to ensure that playing with the Y axis doesn’t make the graph start lying, though.

1 1 5

Danyel Fisher @danyelf.bsky.social · 29d

If I heard it right, the student said in the video she’s there as an “observer.” She’s not a student — she’s volunteer thought police.

1 3

Reposted by Danyel Fisher

Dr. Damien P. Williams Wants A Better Sociotechnical Paradigm @wolvendamien.bsky.social · Sep 8

How many times, in how many contexts, from how many internal and external researchers, or from how many CEO's are people going to have to receive this message before they believe it:

"Hallucinations" are an inherent part of the large language model architecture.
www.nature.com/articles/d41...

Can researchers stop AI making up citations?

OpenAI’s GPT-5 hallucinates less than previous models do, but cutting hallucination completely might prove impossible.

www.nature.com

18 200 650

Danyel Fisher @danyelf.bsky.social · Sep 5

This is sort of the premise of “a Different Flesh” by Harry Turtledove, where North America is inhabited by easily-enslaved evolved Erectus.

6

Danyel Fisher @danyelf.bsky.social · Sep 5

I’m fascinated by all parts of this story. I’m curious what really WAS in the barrels, too — it feels like there’s some story not being told. (Eg the barrels were filled with some other substance, but the prosecutors charged on the heroin.)

2

Danyel Fisher @danyelf.bsky.social · Sep 3

Absolutely. But lots of UI side auto instrumentation gives you that, and so vendors will offer cardinality control because it’s easier than telling people to fix their instrumentation

1 2

Danyel Fisher @danyelf.bsky.social · Sep 3

Sometimes — not always— high cardinality disguises a lack of clear thinking about your data.

Says a man who spent the morning looking at logs that have entries like “user clicked ‘load 37452 rows’” and “user clicked ‘remove db21.s3.x42’

1 1 3

Danyel Fisher @danyelf.bsky.social · Sep 1

Right. “The DOJ Civil Rights Division has determined that mailed in ballots are fundamentally insecure and thus null and void; any elector or representative claiming a seat based on mail in ballots may not be seated.”

1

Danyel Fisher @danyelf.bsky.social · Aug 28

And what municipal architecture was at Cambridge!

1 1

Danyel Fisher @danyelf.bsky.social · Aug 28

That if these walls could talk, that doesn’t mean they’d be thinking, and they’d mostly mutter about paint in run-on sentences.

1

Danyel Fisher @danyelf.bsky.social · Aug 27

What can we do to help get you started? An unopposed seat tells people they have no hope. Having Bethany running becomes the first step towards letting people express themselves (and forces your opponent to actually think about what he’s saying)

Danyel Fisher @danyelf.bsky.social · Aug 27

But the more important thing: we don't need to either accept the harms are inevitable, or try to criminalize technology. We just need to (a) stop shoving a large chunk of GDP into making things actively worse, and (b) allow liability to flow in the normal way.

1

Danyel Fisher @danyelf.bsky.social · Aug 27

I'd disagree: Facebook went *very quickly* from "chat with your high school buddies" to "oops, did we trigger a massacre in Ethiopia?", also with plenty of warning from clear-eyed people who were watching it happen, and a billionaire who didn't want to f** with profits.

www.cnn.com/2021/10/25/b...

Facebook knew it was being used to incite violence in Ethiopia. It did little to stop the spread, documents show | CNN Business

The social media giant has been used by militia groups and political leaders to wage a virtual battle amid the Tigray war.

www.cnn.com

1 1

Danyel Fisher @danyelf.bsky.social · Aug 27

Sure. Harmful technologies exist and have existed. Does it change anything when there is a legal chain of liability for using, installing, distributing, and training those models?

1

Danyel Fisher @danyelf.bsky.social · Aug 27

And yet people subscribe to Netflix and buy books and get their prescriptions drugs.

This isn’t a conversation of “can we ensure there will be zero ai in the world.” It’s a conversation of “maybe we don’t spend the GDP on a dangerous technology without some controls”

2 3

Danyel Fisher @danyelf.bsky.social · Aug 27

(And legally dangerous. Does Microsoft sink billions of dollars into AI if they know they’re on the hook for telling teenagers to kill themselves?)

Danyel Fisher @danyelf.bsky.social · Aug 27

I see an important difference between “our national policy is to spend as much money on AI as possible” and “we can’t stop all of it but we can make it expensive and unpopular”

1 2

Danyel Fisher @danyelf.bsky.social · Aug 27

And yet despite the fact that not all countries are on board with it, pushing things like IP piracy and online contraband distribution underground has successfully slowed those industries.

1 2

Danyel Fisher @danyelf.bsky.social · Aug 27

Allow lawsuits or criminal charges against AI hosts, data centers, trainers, and operators for damage, IP theft, or generated hate speech. You don’t need to ban it — just make it too expensive to use.

2 2

Danyel Fisher @danyelf.bsky.social · Aug 24

I couldn’t disagree more — your research seems incredibly relevant to the current time.

1 1

Danyel Fisher @danyelf.bsky.social · Aug 24

Bethany for Rep! Run Bethany Run!

2 2

Danyel Fisher @danyelf.bsky.social · Aug 20

That’s ok. We will paint it white every evening so we can see dark-dressed figures climbing it in the dark. Two paint layers a day