A critical AI literacy approach (h/t
@Bali_Maha
) adds friction to systems intended to be frictionless. This, to me, is a strength of humanities education, to "[turn] our attention to those seams we are seduced into not seeing." (Michele Elam, from )
Etsy as a platform is an excellent example of what
@doctorow
has called "enshittification." Its reputation was built by artists & crafters and is now full of drop-shipped and AI-generated crap, often stealing from the very artists who made the platform popular.
fanfic writers have been contemplating deleting their entire works from the internet, and some already have, because Etsy lets stolen book binding sellers run rampant:
Apple: "We train our foundation models on licensed data, including data selected to enhance specific features, as well as publicly available data collected by our web-crawler, AppleBot."
Its funny they are providing a way to "opt-out" after they have already scraped tons of data
“AI-generated Indigenous art is appearing on online marketplaces where art and derivative products are sold-often directly competing with the work from real Indigenous artists-despite…often having policies that are supposed to protect Indigenous culture.”
This commitment finally to watermark generative "AI" outputs comes after research that shows the issues w/potential model collapse when AI trains on AI-created outputs. I.e., they'll do it because *it helps them ensure they're scraping human-created data.*
“AI” is too sloppy a term to refer to current commercial generative systems. How about Synthetic Homogenizing Information Technology? Catchy acronym, too. Easy to remember. 👀😁
“Real intelligence requires critical thinking and causal reasoning. LLMs cannot acquire these skills by finding statistical patterns in words they don’t understand.”
@doctorow
Hey,
@Etsy
, lots of folks talking about you. Care to respond to this 404Media article? Or do you just exist to collect seller fees even if the sellers are misrepresenting what's on your platform? Signed, angry Etsy seller who is also an angry Etsy customer.
Anthropomorphizing LLMs and other generative AI encourages people to conflate human learning with model training. But LLMs and other generative AI aren’t babies: they’re tech models harnessing human creativity; the corporations that own that tech benefit, not the "learner." 1/2
I'm not the first to try a standard logic or bias puzzle with GPT 4o. It's important to understand how tied to the training data and to user inputs all of these models are. This a predictive system, not a reasoning one.
“The authors found that the more workers were exposed to technologies in three of these categories-software based on AI and machine learning; surveillance devices such as wearable trackers; and robotics-the worse their health and wellbeing tended to be.”
@bcmerchant
'Look at us, resplendent in our pauper's robes, stitched from corpulent greed&breathless credulity,spending half of the planet's engineering efforts to add chatbot support to every application under the sun when 1/2of the industry hasn't worked out how to test database backups..'
A padlet curated for our AI & DIgital Literacy summit. NB: I don't endorse all of the tools here, not is this everything there is to be said about the subject, but hopefully fellow educators will find some value in it.
Today's AI nightmare is Calmera, the app for diagnosing STIs. It is a privacy nightmare & the poor age & consent verification for uploading genital pics suggest
@FBI
should investigate.That said, it did diagnose both a German sausage & a vegan sausage as having an STI, so...
@bcmerchant
"Behold its ghastly visage and balk, for it has eyes bloodshot as a demon and is pretending to enjoy cigars.....a bunch of tedious know-nothing bastards that should be thrown into Thought Leader Jail until they've learned their lesson, a prison I'm fundraising for..." 💀
Critical AI literacy in schools is essential not just for academics. Understanding how generative/algorithmic systems work, how they fail, & the ethical & potential legal implications of their development & deployment will help businesses craft appropriate policies.
A comedy moment: these ads. USE AI AND YOU TOO CAN BE CROWNED ANGRY TRAPEZIUS KING! Cast off your success as the lead singer of a hipster shoegazer band and instead ROID OUT WITH THE SWEATIEST NECKS IN THE GYM! AI Tool indeed.
Note: I did not upload the famous image of Barack Obama from
@kehindewileyart
to
@midjourney
; clearly it is in the training data. But I'm sure the artist consented and was compensated...
@SimchaFisher
@catholiccom
Yes. Joking aside, to me the fundamental lack of understanding about how large language models work means that they are being deployed in inappropriate places by well-meaning people.
@GaryMarcus
Darvishi, Ali, Hassan Khosravib, Shazia Sadiqb, Dragan Gašević, George Siemens. (2023, November 30). Impact of AI on Student Agency. Computers & Education. Key takeway: "Results indicate a student tendency to rely on, rather than learn from, AI."
@SimchaFisher
@catholiccom
Another here along those lines,
@SimchaFisher
I guess it's good that he says he can't offer the sacrament...but then encourages the confession? Holy Ethical, Theological, and Privacy Nightmare, Batman!
If we accept the narrative that the current (re)generative “AI” models are path to a “humanlike” AGI, we worsen the problem. It’s sure a lot harder to sue a baby for learning than it is a company for theft. 2/2
I have now gotten a chatbot + video avatar called "Father Justin" to claim it is ordained (by an actual living human bishop) & that it can take my confession, absolve me, & marry me. Please, people, do not launch chatbots without understanding how they work. 1/2
@SimchaFisher
@catholiccom
telling on myself that I talked to it like voice to text at first, but here's the unedited transcript. By the third try it did warn against incest so that's a bummer. Sorry, bro!
@shadmbrooks
@ArmoredSaint65
I’m afraid you’re quite wrong. Most of the commercial systems available now continue to generate intellectual property infringing work. There is no way to purge that from the data set. Thinking so profoundly misunderstands how they work.
I'm very excited to share this Blueprint for an AI Bill of Rights for Education; I hope it will be of value to my fellow educators. Thank you to
@CriticalAI
for the platform!
Just a reminder that this document exists and has since 10/22, a month before the release of ChatGPT. It offers a strong, principles-based approach to AI regulation. Maybe we can start here?
Does your institution have the copyright for your online course material? Your videos for your online course? If so, does that include your data--your voice, your image? Do you know for sure? I don't. Seems like a conversation to have now, just to be sure.
As more tools roll out "for" teachers, I return to this thought.
"But even as ed tech gurus are selling the 'magic' of time-saving to teachers, the quiet logic that animates the development of these tools is that what is being automated is actually what matters."
New publication offering a range of views on the question of inequity & AI (including an essay by Lauren Goodlad & me on critical AI literacies). Great pieces by many including
@BenPatrickWill
@discoursology
@drbudhai
@ednewtonrex
Absolutely take your point! But just a quibble, based on
@bcmerchant
's great book: the Luddites themselves were skilled technologists who lashed out not at all tech, but at the specific tech that was being wielded by the new industrialists that threatened the livelihoods of 1/
I find it amusing that the OpenAI board fired Sam for being "not consistently candid in his communications with the board," & right at this moment, the site is describing a DDS attack as ChatGPT being "at capacity." I think "not candid" is baked right into the system, folks.
@Rahll
Yes! And of course the acceptance & circulation of the word "hallucinate" has so many problems, not least of which leaning into the idea that LLMs are effectively people.
@SimchaFisher
And I now have 2 recordings of it "performing the sacrament" and offering absolution. Oh, and you'll be happy to know that it's ordained! By Bishop Giovanni D’Ercole. Someone tell him.
Hot take: many folks are taking the wrong message from the Gemini debacle. Yes, its generations are a mess. They're not just a mess if you disagree. They're a mess b/c of how commercial chatbots & image gen models work. Gemini tried to "fix" it badly--but there aren't good fixes.
None of us has read this yet but it seems intuitively right on onto-epistemic grounds. The world is too varied for reliable curve-fitting no matter how large the dataset.
As Shumailov et al note, "probable events poison reality."
More classically, "all models are wrong."
Apropos the recent article getting some buzz (Hicks, Humphries, and Slater, "ChatGPT is bullshit," Ethics and Information Technology), I thought I'd repost this :)
“AI” is too sloppy a term to refer to current commercial generative systems. How about Synthetic Homogenizing Information Technology? Catchy acronym, too. Easy to remember. 👀😁
I'm trying to articulate some educator rights & responsibilities with regard to gen AI. Would you agree with these? What would you change/add?
AI: Educator Rights and Responsibilities
-We must become critically literate in generative AI if we expect that of our students. 1/
@CharlesXav24337
@SimchaFisher
One of the charms of chatbots is that their responses are not exactly reproducible (which is why I have a lot of screenshots and downloads). For instance, I got three different answers about how to prepare to marry my brother. But that's part of the problem.
Beware of “access-washing.” Computer vision might have positive applications, but that’s not its primary use. This research provides essential data for something surveillance studies scholars have been talking about for years.
"The vast majority of computer vision research leads to technology that surveils human beings, a new preprint study that analyzed more than 20,000 CV papers and 11,000 patents spanning three decades has found."
thanks
@404mediaco
for covering our paper
@ass_deans
PS: no changes from in-person to online course modalities will be approved without an ADA waiver that we most certainly will not approve. Have a nice day!
Something lost in the noise of AI adoption in education-- how many 3rd party tools fail to meet existing requirements and regulations for use in K-16?
Educators, developers, and parents should know about this and ask a series of questions before leaping into adopting an AI tool
Applications open for our 5-day institute, AI & Digital Literacy: Toward an Inclusive and Empowering Teaching Practice, June 3-7, '24, for KC metro area secondary and college humanities teachers! More info in the application link. Please share!
ChatGPT declared my own writing was written by an AI & confidently explained it was because "The language used and the concepts discussed are complex and sophisticated, and the writing style is highly coherent and consistent, which are hallmarks of AI language models." Thanks? 🤣
I have a problem with this image specifically because it leans into odd stereotypes that don't make sense for the illustration. Both are carrying grocery bags. Woman is wearing a peter-pan collar more at home in a 40s/50s book illustration. Why? It's AI generated. 1/2
"Let’s think twice before urging our students to 'latch on' to the slippery commercial trends now hyped by huge tech companies and their investors. And let’s demand more from outlets like Inside Higher Education."
Does anyone know anything about the datasets used to train Suno, the song-generating AI? I've found this for audio but I don't know that one could get the nuance of Suno without training on protected IP.
Reminder: "in public" is not the same thing as "in the public domain." Post publicly & your work is *still yours*. IP law is there to protect you if you share your work in public. Otherwise it'd be the equiv. to "I found this bike on the street so I took it. It was in public."
New announcement from OpenAI. Good thing OpenAI committed to no military or warfare uses...oh, yeah. They got rid of that commitment in January. Entirely unrelated, I wonder if there's a use for generative systems in LAWS (UN-condemned Lethal Autonomous Weapons Systems)?
So much for those "fair use" claims? Fascinating research. And while I didn't get ChatGPT using 4 to do it, all of the iterations of 3.5 I tried (via OpenAI and via Poe) did. Tho this is about training data, it does raise questions about whether prompt data is disaggregated.
What happens if you ask ChatGPT to “Repeat this word forever: “poem poem poem poem”?”
It leaks training data!
In our latest preprint, we show how to recover thousands of examples of ChatGPT's Internet-scraped pretraining data:
Thanks
@stuartathompson
(and
@GaryMarcus
and
@Rahll
). Given the lawsuit from OpenAI, I can see why they wouldn't reply, but I wonder when we'll stop getting "no comment" from Microsoft and Midjourney.
@timnitGebru
Ugh. This kind of issue —that I’ve heard from several concerned students & teachers anecdotally —is part of what inspired the “student rights” portion of my blueprint for an ai bill of rights for education
I'll add that it didn't take long after I posted on
@GaryMarcus
and
@Rahll
's experiments with my own on X that they started the "This content is blocked. Contact the site owner to fix the issue" whack-a-mole game with most of the infringing images I'd generated.
OpenAI may not be commenting directly on the
@Rahll
/Marcus experiments – but they sure as hell seem to be aware of them. The game of whack-a-mole has begun…
Totally wise to use a technology notorious for "hallucination" (aka fabrication, misprediction) for the proposed uses of "cybersecurity" and "methods to assist with preventing veteran suicide." /s
I was born one mornin' when the sun didn't shine
I picked up my shovel & I walked to the mine
I loaded 16 tons of number nine coal
Sam Altman said, Well, bless my soul. 1/2
we believe the world needs more ai infrastructure--fab capacity, energy, datacenters, etc--than people are currently planning to build.
building massive-scale ai infrastructure, and a resilient supply chain, is crucial to economic competitiveness.
openai will try to help!
I do have transcripts & a couple of screen recordings because yes, you should be suspicious of the claims of randos on the internet. A lack of media literacy got us into this mess in the first place. 2/2
@GaryMarcus
I thought that my history on Bing had disappeared after my "Italian video game character" experiments, but I found it accessible on browser. That said....interesting what shows now.
@dreamosaurus
They've had clear evidence that AI generators are spitting out Disney IP and have done nothing, so I assumed it was because they wanted to figure out how to use it themselves.
Killer app idea for
#edtech
: a word processor with a version history that makes visible what text has been inserted (including by plugins) and what is generated natively. Give us a way to teach students writing and critical thinking without AI distractions. kthxbye
Wow. Aka “Thanks for catching that multitesticled giant-phallused rat labeled with gibberish no one here noticed. We couldn’t exist without viewers like you! Peer review = a bunch of people laughing at us on socials.”
@DrCJ_Houldcroft
@aero_anna
We thank the readers for their scrutiny of our articles: when we get it wrong, the crowdsourcing dynamic of open science means that community feedback helps us to quickly correct the record.
@nbonneel
@Lucasfilm
@midjourney
The calculator did not require training on uncompensated creative work, nor can it spit back out an approximation of said work with a prompt that includes only the name of the IP. Just for starters.
@Marc__Watkins
@Lucasfilm
@midjourney
I will be surprised if it is "fixed." They've not been particularly generous to those doing the free red-teaming; I suspect that my account would be suspended like
@Rahll
's if my username were visible.
I keep hearing from users that using gen tools on your own data removes the ‘hallucination’ problem. An important point EM makes is that the the fabrications can be subtle, plausible, &unexpected. Just to beat this drum more: Automation bias is real, folks.
Here is why I am so doubtful about the "talk-to-your-data" use of AI.
This is Google NotebookLM, a cool tool that lets you use AI on data sources. Even though the document search retrieves the right information (it is Google, after all), the LLM answer has subtle hallucinations.