#chatgpt

[kareguya]

Umaima

[animentality]

[memendoemori]

Good morning everybody

[softwaring]

this reply kills me 😭 article link

[thehungrycity]

[dduane]

Well, this would be interesting...

[purpleartrowboat]

ai makes everything so boring. deepfakes will never be as funny as clipping together presidential speeches. ai covers will never be as funny as imitating the character. ai art will never be as good as art drawn by humans. ai chats will never be as good as roleplaying with other people. ai writing will never be as good as real authors

[brookheimer]

favorite thing ab chatgpt is that if it doesn’t know something it’ll just start fucking lying. like blatantly fucking lying.

my dad teaches english classes and he just got a final paper with this sentence: “In terms of style, both poets are known for their use of imagery, but O'Hara's tends to be more straightforward and concrete, while Stevens' is often more abstract and metaphorical — for example, in O'Hara's poem "The French / Window," he writes: "A cat walks along the garden wall / and the tree waves its branches / The French / windows are blah" (lines 1-4).”

the thing about “The French / Window” is that it is not a poem that exists. at all. like, it was literally just written by chatgpt then inexplicably named as a famous frank o’hara poem. and it’s so. fucking. funny. sooo basically heads up for finals season — those of you who use chatgpt, be warned, because you will quite literally be citing nonexistent texts and your professors will show it to their daughters and together they will laugh at you endlessly and you will deserve it

[postmodernmulticoloredcloak]

anyway in case you don’t know it yet

[porcupine-girl]

An important message to college students: Why you shouldn't use ChatGPT or other "AI" to write papers.

Here's the thing: Unlike plagiarism, where I can always find the exact source a student used, it's difficult to impossible to prove that a student used ChatGPT to write their paper. Which means I have to grade it as though the student wrote it.

So if your professor can't prove it, why shouldn't you use it?

Well, first off, it doesn't write good papers. Grading them as if the student did write it themself, so far I've given GPT-enhanced papers two Ds and an F.

If you're unlucky enough to get a professor like me, they've designed their assignments to be hard to plagiarize, which means they'll also be hard to get "AI" to write well. To get a good paper out of ChatGPT for my class, you'd have to write a prompt that's so long, with so many specifics, that you might as well just write the paper yourself.

ChatGPT absolutely loves to make broad, vague statements about, for example, what topics a book covers. Sadly for my students, I ask for specific examples from the book, and it's not so good at that. Nor is it good at explaining exactly why that example is connected to a concept from class. To get a good paper out of it, you'd have to have already identified the concepts you want to discuss and the relevant examples, and quite honestly if you can do that it'll be easier to write your own paper than to coax ChatGPT to write a decent paper.

The second reason you shouldn't do it?

IT WILL PUT YOUR PROFESSOR IN A REALLY FUCKING BAD MOOD. WHEN I'M IN A BAD MOOD I AM NOT GOING TO BE GENEROUS WITH MY GRADING.

I can't prove it's written by ChatGPT, but I can tell. It does not write like a college freshman. It writes like a professional copywriter churning out articles for a content farm. And much like a large language model, the more papers written by it I see, the better I get at identifying it, because it turns out there are certain phrases it really, really likes using.

Once I think you're using ChatGPT I will be extremely annoyed while I grade your paper. I will grade it as if you wrote it, but I will not grade it generously. I will not give you the benefit of the doubt if I'm not sure whether you understood a concept or not. I will not squint and try to understand how you thought two things are connected that I do not think are connected.

Moreover, I will continue to not feel generous when calculating your final grade for the class. Usually, if someone has been coming to class regularly all semester, turned things in on time, etc, then I might be willing to give them a tiny bit of help - round a 79.3% up to a B-, say. If you get a 79.3%, you will get your C+ and you'd better be thankful for it, because if you try to complain or claim you weren't using AI, I'll be letting the college's academic disciplinary committee decide what grade you should get.

Eventually my school will probably write actual guidelines for me to follow when I suspect use of AI, but for now, it's the wild west and it is in your best interest to avoid a showdown with me.

[hashtagloveloses]

the US Copyright Office is asking for public comment on AI and copyright!!! now is the time to be LOUD about it so there can be government regulations.

GO COMMENT RIGHT NOW:

[nobodysuspectsthebutterfly]

FYI artists and writers: some info regarding tumblr's new "third-party sharing" (aka selling your content to OpenAI and Midjourney)

You may have already seen the post by @staff regarding third-party sharing and how to opt out. You may have also already seen various news articles discussing the matter.

But here's a little further clarity re some questions I had, and you may too. Caveat: Not all of this is on official tumblr pages, so it's possible things may change.

(1) "I heard they already have access to my data and it doesn't really matter if I opt out"

From the 404 article:

A new FAQ section we reviewed is titled “What happens when you opt out?” states “If you opt out from the start, we will block crawlers from accessing your content by adding your site on a disallowed list. If you change your mind later, we also plan to update any partners about people who newly opt-out and ask that their content be removed from past sources and future training.”

So please, go click that opt-out button.

(2) Some future user: "I've been away from tumblr for months, and I just heard about all this. I didn't opt out before, so does it make a difference anymore?"

Another internal document shows that, on February 23, an employee asked in a staff-only thread, “Do we have assurances that if a user opts out of their data being shared with third parties that our existing data partners will be notified of such a change and remove their data?” Andrew Spittle, Automattic’s head of AI replied: “We will notify existing partners on a regular basis about anyone who's opted out since the last time we provided a list. I want this to be an ongoing process where we regularly advocate for past content to be excluded based on current preferences. We will ask that content be deleted and removed from any future training runs. I believe partners will honor this based on our conversations with them to this point. I don't think they gain much overall by retaining it.”

It should make a difference! Go click that button.

(3) "I opted out, but my art posts have been reblogged by so many people, and I don't know if they all opted out. What does that mean for my stuff?"

This answer is actually on the support page for the toggle:

This option will prevent your blog's content, even when reblogged, from being shared with our licensed network of content and research partners, including those that train AI models.

And some further clarification by the COO and a product manager:

zingring: A couple people from work have reached out to let me know that yes, it applies to reblogs of "don't scrape" content. If you opt out, your content is opted out, even in reblog form. cyle: yep, for reblogs, we're taking it so far as "if anybody in the reblog trail has opted out, all of the content in that reblog will be opted out", when a reblog could be scraped/shared.

So not only your reblogged posts, but anyone who contributed in a reblog (such as posts where someone has been inspired to draw fanart of the OP) will presumably be protected by your opt-out. (A good reason to opt out even if you yourself are not a creator.)

Furthermore, if you the OP were offline and didn't know about the opt-out, if someone contributed to a reblog and they are opted out, then your original work is also protected. (Which makes it very tempting to contribute "scrapeable content" now whenever I reblog from an abandoned/disused blog...)

(4) "What about deleted blogs? They can't opt out!"

I was told by someone (not official) that he read "deleted blogs are all opted-out by default". However, he didn't recall the source, and I can't find it, so I can't guarantee that info. If I get more details - like if/when tumblr puts up that FAQ as reported in the 404 article - I will add it here as soon as I can.

Edit, tumblr has updated their help page for the option to opt-out of third-party sharing! It now states:

The content which will not be shared with our licensed network of content and research partners, including those that train AI models, includes: • Posts and reblogs of posts from blogs who have enabled the "Prevent third-party sharing" option. • Posts and reblogs of posts from deleted blogs. • Posts and reblogs of posts from password-protected blogs. • Posts and reblogs of posts from explicit blogs. • Posts and reblogs of posts from suspended/deactivated blogs. • Private posts. • Drafts. • Messages. • Asks and submissions which have not been publicly posted. • Post+ subscriber-only posts. • Explicit posts.

So no need to worry about your old deleted blogs that still have reblogs floating around. *\o/*

But for your existing blogs, please use the opt out option. And a reminder of how to opt out, under the cut:

The opt-out toggle is in Blog Settings, and please note you need to do it for each one of your blogs / sideblogs.

On dashboard, the toggle is at https://www.tumblr.com/settings/blog/blogname [replace "blogname" as applicable] down by Visibility:

For mobile, you need the most recent update of the app. (Android version 33.4.1.100, iOs version 33.4.) Then go to your blog tab (the little person icon), and then the gear icon for Settings, then click Visibility.

Again, if you have a sideblog, go back to the blog tab, switch to it, and go to settings again. Repeat as necessary.

If you do not have access to the newest version of the app for whatever reason, you can also log into tumblr in your mobile browser. Same URL as per desktop above, same location.

Note you do not need to change settings in both desktop and the app, just one is fine.

I hope this helps!

[noosphe-re]

"There was an exchange on Twitter a while back where someone said, ‘What is artificial intelligence?' And someone else said, 'A poor choice of words in 1954'," he says. "And, you know, they’re right. I think that if we had chosen a different phrase for it, back in the '50s, we might have avoided a lot of the confusion that we're having now." So if he had to invent a term, what would it be? His answer is instant: applied statistics. "It's genuinely amazing that...these sorts of things can be extracted from a statistical analysis of a large body of text," he says. But, in his view, that doesn't make the tools intelligent. Applied statistics is a far more precise descriptor, "but no one wants to use that term, because it's not as sexy".

'The machines we have now are not conscious', Lunch with the FT, Ted Chiang, by Madhumita Murgia, 3 June/4 June 2023

[aiweirdness]

Original

It had generated a complete reference with title, author, and everything. Completely fabricated.

It’s not a search engine, it’s a crappy imitation of one.

[kenyatta]

I once had ChatGPT insist that a particular composer wrote music for a game, even going so far as to list particular songs from the soundtrack that they were supposedly responsible for, and it helpfully provided hallucinatory citations when I asked for them (a broken link on the game publisher's website and a link to Wikipedia, which did not in fact support its assertion either now or at any point in the article's history). Nor could I find anywhere else on the internet where someone even mistakenly believed that that composer had worked on the game. ChatGPT lies not because it's regurgitating falsehoods that it found on the internet - it lies because it invents new falsehoods on its own. It's not just trained on stuff on the internet that's wrong; it's trained to be confidently wrong in general. It doesn't know what facts are, it just knows how to produce things that are shaped like facts and shove them in fact-shaped holes. I personally wasted 30 minutes of my life fact-checking/"not believing everything it says", when it confidently told me something surprising. My horizons were not broadened by exposing me to "different worldviews". This was unequivocally a negative experience for me.

comment on a MetaFilter post about AI: "My goal is to be helpful, harmless, and honest."

[shortace]

Just on a whim, because I know that Alcibiades is one of the weirdest and funniest characters in ancient Greek history, I asked ChatGPT "What's the weirdest thing Alcibiades ever did?"

ChatGPT came back with the details of something Alcibiades (henceforth referred to as 'Alci' so I don't have to keep typing it out) was accused of, but acquitted of.

When I pointed out that he had been acquitted and may not have actually done this thing, Chat GPT apologised and said, "yes, he was acquitted", and then went on to tell me that, nonetheless, the event was significant because it made Alci flee the city.

Alci did not flee the city, he was sent away on a military expedition, which was exactly what he'd wanted and asked for. When I pointed that out, ChatGPT apologised again for being wrong.

I asked again for weird things he might actually have done, and was told one version of a story I've heard before about how Alci stole some stuff from a friend. ChatGPT's version was different from what I'd heard, though, so I mentioned that, and only then did ChatGPT acknowledge that there were different versions of the story. As part of its apology and correction, ChatGPT said that it did not always have access to all information - but then proceeded to provide details of the version of the story I'd heard before, showing that it did, in fact, have access to that information.

I asked again, what is the weirdest thing Alcibiades ever did? ChatGPT gave me an answer, which was a story I'd never heard before, so I asked for a source. ChatGPT told me it was in Plutarch's Lives, and I presumed it was in his Life of Alcibiades, so that's where I looked. When I said I couldn't find it there, ChatGPT told me, sorry for not being specific, it was actually in Plutarch's Life of Nicias. So I went and read Plutarch's Life of Nicias and couldn't find it.

So I told ChatGPT that I couldn't find the story in that book, could it please be more specific? What I was hoping for was a chapter or page number or something, I just presumed I'd missed it.

ChatGPT came back with "no, actually it's not in that book, it may be a later invention, there is no concrete evidence for this story."

TL;DR: ChatGPT cannot be trusted. Even when it does give you a source, it can be wrong. It has no capacity to evaluate the accuracy or likely accuracy of the information it gives you. It will present you with wrong or debatable information and give you absolutely no indication that it may not be correct, or that other versions or interpretations are possible.