Lightnews — Scholar-powered news

Reposted by Jessica Price

Dane 🔜 NYCC AA L-34 @monkeyminion.com · 10h

I wrote a 15 page report on heraldic symbolism in medieval armor and weapon design for my art history class the night before it was due (8am class). Made up 90% of it (only found one book for reference) and got an A. GenAI could fucking never.

wrote 20 pages on Faulkner's The Bear four hours before final papers were due on trucker pills and coffee and cigarettes and got an A, fuck you.
You people couldn't hang with real slackers.
finn
wokeupchic • 4d
It's fuck Al till your homework due in 25 minutes

7 170 1.2K

Jessica Price @delafina777.bsky.social · 6h

I wrote a 20-page paper in an hour and a half to get it turned in on time. Got an A and a thank you from the prof for it being fun to read.

Relying on AI makes you weak. You will not survive the winter.

Dane 🔜 NYCC AA L-34 @monkeyminion.com · 10h

I wrote a 15 page report on heraldic symbolism in medieval armor and weapon design for my art history class the night before it was due (8am class). Made up 90% of it (only found one book for reference) and got an A. GenAI could fucking never.

7 61

Jessica Price @delafina777.bsky.social · 6h

Well, I now have 379 responses and the results have changed with a larger sample size. (For one thing, this morning, with about half that number, it looked like people who read 6+ books a month actually performed the worst, and people who read 0-2 the best. 6+ are now performing the best.

Jessica Price @delafina777.bsky.social · 1d

Hey all, I'm trying to get a baseline for how good people are at detecting whether fiction is AI-generated or human-written (most of the research on this is for nonfiction and my hypothesis is we're better with fiction).

If you have time, give it a try?

jlp-experiment.github.io/ai-detection...

AI vs Human Fiction Detection Test

jlp-experiment.github.io

2 6 14

Reposted by Jessica Price

This Is Completely Unprepared, and I Am Real @jsx.bsky.social · 17h

'“Yesterday morning, I was broken into again for the tenth time,” says a business owner in downtown Portland.'

I checked on this - the business owner is Amy Nichols, a Republican running for office in Clackamas County who withdrew her sponsorship of the Portland Pickles over LGBTQ representation

1 11 79

Reposted by Jessica Price

eric shamow @botmatrix.myatproto.social · 15h

If you've ever been curious if the concept of a Trust Thermocline was real, the change in tenor and good faith in the relationships on this site over the past week should be pretty revealing.

every.to/p/breaching-...

Breaching the Trust Thermocline Is the Biggest Hidden Risk in Business

Companies have no way back from a sudden loss of trust

every.to

6 15

Jessica Price @delafina777.bsky.social · 13h

Participants got randomly assigned to a control or experimental group. The experimental group gets shown a labeled AI and labeled human sample to prime them, although it doesn’t seem to have made much of a difference in the results.

Jessica Price @delafina777.bsky.social · 13h

Ahhh ok, that makes more sense. Yeah, it’s feedback I’ve heard from multiple people. I added a note to the instructions that there’s a chance the samples will be all AI or all human.

1

Jessica Price @delafina777.bsky.social · 14h

Wait, you saw the exact same sample 3 times?

1

Jessica Price @delafina777.bsky.social · 15h

The text doesn’t really say. Christians have assumed it was because they were social outcasts, but I also think it’s worth noting that he was surviving via patronage and while they probably didn’t have nobility-money, they likely had money.

2

Jessica Price @delafina777.bsky.social · 15h

Well, no, unfortunately, I classified human-written, AI-polished samples as human.

It might have been AI with a robust prompt (stylistic info, outline, human examples), though.

1

Jessica Price @delafina777.bsky.social · 15h

The poster this is quoting disabled quote tweets, but let me say this in response to it:

purity tests are things like “are you the RIGHT kind of socialist?”

not “are you willing to let the government murder some marginalized people? how many?”

keith kurson @keith.is · 20h

If you're willing to sacrifice the groups with the least power to protect yourself, I'm going to make the assumption that it'll be easy for you to sacrifice every group besides your own.

This is the only hill to die on.

the tallest stone @talleststone.bsky.social · 22h

I hate to say it but the majority of the country is not going to die on the hill of trans rights. We can’t protect ANY groups if we can’t win a national election, so purity tests only serve to make specific groups feel triumphant. That’s not “erasing” them; it’s long-horizon strategy.

1 8 44

Jessica Price @delafina777.bsky.social · 15h

Yeah, it also sort of over-formalized things (e.g. changing “stupid girl” to “foolish woman”).

1 4

Jessica Price @delafina777.bsky.social · 16h

And, sometimes, ESL writers.

1 4

Jessica Price @delafina777.bsky.social · 16h

No. I classified them as human or AI based on who or what wrote them.

1

Jessica Price @delafina777.bsky.social · 16h

Ooh, maybe! I wonder if anyone knows anyone there.

1

Jessica Price @delafina777.bsky.social · 16h

Might have been AI-polished—the “polish” weirdly introduced more errors than it caught.

The pure human samples got an edit pass from me a few days after I wrote them, and a second one from another human editor, so I think when I pushed the AI to change things, it didn’t know what to change.

1 3

Jessica Price @delafina777.bsky.social · 16h

I would really love to partner with a university or something to test this properly.

1 1

Jessica Price @delafina777.bsky.social · 16h

I mean, people didn’t reliably FLAG the AI-polished human samples, but they also didn’t LIKE them.

1 1

Jessica Price @delafina777.bsky.social · 16h

(And of course genre is a confounding factor here because I’m not mean enough to ask anyone to read 5 samples in a genre they don’t like.)

1 6

Jessica Price @delafina777.bsky.social · 16h

The other thing that’s interesting is it seems to be showing multiple pure human samples together more often than it should, which suggests the preference for human samples might actually be even stronger than it looks, since they’re competing.

1 9

Jessica Price @delafina777.bsky.social · 16h

PLEASE don't take it more than once. There's a good chance you'll see repeated samples that you already know the answer to, which will skew the results.

2 18

Jessica Price @delafina777.bsky.social · 16h

I did add a note to the instructions letting participants know that the selection of samples they see might be all AI or all human.

2 11

Jessica Price @delafina777.bsky.social · 16h

Wait, please don't take it multiple times! If you see repeated samples, you'll know the answers, and that will skew the results.

1 1

Jessica Price @delafina777.bsky.social · 16h

Please don't! There's a chance you'll see samples you already saw, which would skew the results.

1 1

Jessica Price @delafina777.bsky.social · 16h

Yeah, my hypothesis is that there is actual unconscious expertise here. If you skim, the AI will fool you. If you overthink, the AI will fool you.

But people *liked* the human samples better.

1 3