wendy norris
LightNews LightNews
banner
wendynorris.bsky.social
wendy norris
@wendynorris.bsky.social
1.3K followers 2.4K following 680 posts
Asst Prof | Ethical Data Science
Ex-investigative reporter and editor
Crisis informatics nerd
Makes good trouble
Tell your dog I said "hi"
Posts Replies Media Videos
wendynorris.bsky.social
I have a preprint of a book chapter on post-API social media data collection with a timeline of platforms’ CYA moves due to privacy scandals, lawsuits, and the rise of LLMs.

Happy to send it your way if it would be helpful.
November 21, 2025 at 2:42 PM
wendynorris.bsky.social
New Gmail sig line
Please excuse any typos or delays in responding. Google deleted spellcheck and message filtering in Gmail if you turn off their AI spy now embedded in Gmail, Drive, Docs & more. Here's how to opt-out of Gemini: https://www.zdnet.com/article/how-to-turn-off-gemini-in-your-gmail-photos-chrome-and-more-its-easy-to-opt-out-of-ai/ ..
November 21, 2025 at 3:50 AM
wendynorris.bsky.social
Google Gemini is also in the game spreading its tentacles into all Google Ed products. This popped up in a student's programming project yesterday.

Mere weeks before the end of the semester when everyone is tired and easily seduced by "time savers". I'm going to flip my lid.
Banner on Google Colab: "College students get the Pro plan of Google Gemini and Colab free. Terms Apply. Learn more at [urls]." Oh boy, about those terms ...
November 19, 2025 at 9:22 PM
wendynorris.bsky.social
So much winning in #EdTech I can hardly breathe
Banner alert in Google Colab: College students get the Pro plan of Google Gemini and Colab free. Terms Apply. Learn more. Oh, I bet "terms apply." From article at WindowsCentral.com 11/18/2025: "Microsoft has issued an important warning about its upcoming agentic AI capabilities that are coming soon to Windows 11. In a new support document, the company warns that users should 'only enable this feature if you understand the security implications,' and has confirmed that because of the potential dangers, it'll be off by default." Security for me but not for thee. Nice.
November 19, 2025 at 2:03 PM
wendynorris.bsky.social
/Joyce Carol Oates leaves the chat
November 10, 2025 at 10:39 PM
wendynorris.bsky.social
I absolutely hear your frustration with this. Peer review is hard enough as it is.

Except loading an abstract into a Gen AI app is highly problematic: What might be original work just became part of a training set.

Two things to consider from this excellent paper:

arxiv.org/pdf/2509.05382
Additional user-provided data sources. Developers draw on other sources of user-provided data for training in addition to chat data. Amazon, Google, and OpenAI state that they may train on files that users upload with their prompts, while Microsoft explicitly excludes the contents of uploaded files. Microsoft and Google note that they may train on voice data, and OpenAI, Google, and Microsoft also state that they may train on user-uploaded images, though Microsoft notes that it attempts to de-identify images by removing “metadata or other personal data and blurring images of faces” (Microsoft 2025a). Microsoft and Meta notes that they may use a user’s data from their other products, in combination with input or output data, as part of model training as well. Microsoft’s privacy policy reads: “Microsoft uses data from Bing, MSN, Copilot, and interactions with ads on Microsoft for AI training” (Microsoft 2025a). Meta’s privacy policy states “We also share information with Meta Companies to support innovation. For example, your videos can help train our products to recognize objects, like trees, or activities, like when a dog chases a ball” (Meta 2024), while its generative AI privacy page notes “we also use information shared on Meta products,” including posts, photos, and photo captions, but not private messages (Meta 2025). Despite the use of ambiguous language (e.g., ”may”) to describe their practices, unless explicitly disclaimed, these companies are training on chat conversations that broadly include detailed snapshots of their users’ personal information, as personal content from photos, posts, and private chats with an AI assistant may be included in subsequent training runs. Dataset accountability. User chat inputs alone are not enough to create and sustain frontier models, which is why publicly available internet data has been widely scraped to train them. While there is little transparency about the datasets used to train major companies’ LLMs (Bommasani et al. 2023), most provide notice that personal data likely exists in their training sets. Khan and Hanna (2025) argue that the external collection practices for training data implicate data privacy harms both at the individual as well as the societal level, emanating from the loss of control over personal information when it is scraped, collected, and reused without consent and out of the original context of collection, with no practical way to opt-out. Drawing on the work of Calo (Calo 2011), they argue that “dataset development as a systematic strategy of ingesting personal information” poses two forms of data privacy harms: subjective harms (emanating from unwanted observation) and objective (unanticipated or coerced use of personal information against an individual without first-hand knowledge). Specific privacy concerns with dataset construction include the automation of data collection and the widespread reach of the collection itself, resulting in systematic, mass surveillance of the online sphere across a myriad of contexts that can eventually lead to downstream harms by automated decision-making systems.
November 2, 2025 at 2:53 PM
wendynorris.bsky.social
It's so rainy and blustery outside today we won't be able to recreate this classic Halloween costume.

#MayTheForceBeWithYou
A white Anatolian Shepherd dog wearing a Princess Leia Halloween costume with a Hopi Squash Blossom hair bun wig
October 31, 2025 at 7:20 PM
wendynorris.bsky.social
Unwanted observation +
Coerced use of personal information =
Systematic mass surveillance
Dataset accountability. User chat inputs alone are not enough to create and sustain frontier models, which is why publicly available internet data has been widely scraped to train them. While there is little transparency about the datasets used to train major companies’ LLMs (Bommasani et al. 2023), most provide notice that personal data likely exists in their training sets. Khan and Hanna (2025) argue that the external collection practices for training data impli- cate data privacy harms both at the individual as well as the societal level, emanating from the loss of control over per- sonal information when it is scraped, collected, and reused without consent and out of the original context of collection, with no practical way to opt-out. Drawing on the work of Calo (Calo 2011), they argue that “dataset development as a systematic strategy of ingesting personal information” poses two forms of data privacy harms: subjective harms (ema- nating from unwanted observation) and objective (unantic- ipated or coerced use of personal information against an individual without first-hand knowledge). Specific privacy concerns with dataset construction include the automation of data collection and the widespread reach of the collection itself, resulting in systematic, mass surveillance of the online sphere across a myriad of contexts that can eventually lead to downstream harms by automated decisionmaking systems.
October 31, 2025 at 1:13 AM
wendynorris.bsky.social
Folks, how much more blatant does the surveillance need to be? Gen AI apps collect your data ... beyond your chat session ... often for years. Don't fall for the squishy ToS language "may use."

Read this paper.

The section on data collection of children's chat app use will light your hair on fire.
Additional user-provided data sources. Developers draw on other sources of user-provided data for training in addition to chat data. Amazon, Google, and OpenAI state that they may train on files that users upload with their prompts, while Microsoft explicitly excludes the contents of uploaded files. Microsoft and Google note that they may train on voice data, and OpenAI, Google, and Microsoft also state that they may train on user-uploaded images, though Microsoft notes that it attempts to de-identify images by re- moving “metadata or other personal data and blurring im- ages of faces” (Microsoft 2025a). Microsoft and Meta notes that they may use a user’s data from their other products, in combination with input or output data, as part of model train- ing as well. Microsoft’s privacy policy reads: “Microsoft uses data from Bing, MSN, Copilot, and interactions with ads on Microsoft for AI training” (Microsoft 2025a). Meta’s privacy policy states “We also share information with Meta Companies to support innovation. For example, your videos can help train our products to recognize objects, like trees, or activities, like when a dog chases a ball” (Meta 2024), while its generative AI privacy page notes “we also use infor- mation shared on Meta products,” including posts, photos, and photo captions, but not private messages (Meta 2025). Despite the use of ambiguous language (e.g., ”may”) to describe their practices, unless explicitly disclaimed, these companies are training on chat conversations that broadly include detailed snapshots of their users’ personal informa- tion, as personal content from photos, posts, and private chats with an AI assistant may be included in subsequent training runs.
October 31, 2025 at 1:03 AM
wendynorris.bsky.social
Folks, how much more blatant does the surveillance need to be? Gen AI apps collect your data ... beyond your chat session ... often for years. Don't fall for the squishy ToS language "may use."

Read this paper.

The section on data collection of children's chat app use will light your hair on fire.
Additional user-provided data sources. Developers draw on other sources of user-provided data for training in addition to chat data. Amazon, Google, and OpenAI state that they may train on files that users upload with their prompts, while Microsoft explicitly excludes the contents of uploaded files. Microsoft and Google note that they may train on voice data, and OpenAI, Google, and Microsoft also state that they may train on user-uploaded images, though Microsoft notes that it attempts to de-identify images by re- moving “metadata or other personal data and blurring im- ages of faces” (Microsoft 2025a). Microsoft and Meta notes that they may use a user’s data from their other products, in combination with input or output data, as part of model train- ing as well. Microsoft’s privacy policy reads: “Microsoft uses data from Bing, MSN, Copilot, and interactions with ads on Microsoft for AI training” (Microsoft 2025a). Meta’s privacy policy states “We also share information with Meta Companies to support innovation. For example, your videos can help train our products to recognize objects, like trees, or activities, like when a dog chases a ball” (Meta 2024), while its generative AI privacy page notes “we also use infor- mation shared on Meta products,” including posts, photos, and photo captions, but not private messages (Meta 2025). Despite the use of ambiguous language (e.g., ”may”) to describe their practices, unless explicitly disclaimed, these companies are training on chat conversations that broadly include detailed snapshots of their users’ personal informa- tion, as personal content from photos, posts, and private chats with an AI assistant may be included in subsequent training runs.
October 31, 2025 at 12:34 AM
wendynorris.bsky.social
This. This right here.
For a time, this vision made sense, even if it was imperfectly implemented. But as with many well-intentioned revolutions, openness has more recently been weaponised. What began as a movement to democratise knowledge has instead become justification for a new kind of extraction — this time not of oil or minerals, but of meaning. This phenomenon has become especially evident with the rise of generative AI, which relies on its voracious appetite for public data to train its models and refine its predictions. [highlighted] In the process, the very datasets, research repositories, and public web archives that were designed to serve the public interest have been harvested to train the large language models now controlled by a few corporations in a handful of countries. The situation is dire but it is not hopeless. In what follows, we describe the problem in greater detail, outline the insufficiency of current mechanisms, and then discuss some possible mitigating responses. [/highlighted]
October 26, 2025 at 3:28 PM
wendynorris.bsky.social
I guess I’ve been very fortunate. Both my rescue pups (a lab mix and an Anatolian Shepherd mix) that both came from really awful situations are excellent judges of character
Anatolian Shepherd dog sitting next to a purple rhododendron bush in bloom
October 15, 2025 at 2:21 PM
wendynorris.bsky.social
Grandmaster Flash has been adopted!

His new family named him "Stilgar" because he has strong Javier Bardem energy*

*please accept this brief moment of silliness to help modulate the authoritarianism in plain sight. We need you in the fight. Find joy wherever you can to keep your spirits up
Side-by-side photos, on left, of a brown tabby cat named Grandmaster Flash with an intense look on his face next to an image, on right, from the movie "Dune" of actor Javier Bardem with an intense look on his face playing the character Stilgar. They could be twins!
October 12, 2025 at 2:25 PM
wendynorris.bsky.social
Image of Excel spreadsheet with Copilot AI insertion in a cell Comments: @mcauliffewithaz: [slaps roof] this billion dollar calculator is so bad at math we have to warn you not to use it for math @hypervisible: “Microsoft also warns against using the AI function for numerical calculations or in ‘high-stakes scenarios’ with legal, regulatory, and compliance implications, as COPILOT ‘can give incorrect responses.’”
October 1, 2025 at 5:32 PM
wendynorris.bsky.social
#RochesterNY friends, meet life of the party Grandmaster Flash. This funny, chatty, OG record-scratching stray boy is fully vetted and ready for his forever home. A kitty-shaped hip hop legend loves cats and dogs too.

#Caturday
Grandmaster Flash in all his grumpy-faced tabby cat glory. Four cats playing with a battery-operated feather toy. Best $29 I ever spent. Grandmaster Flash chilling at bottom right.
September 27, 2025 at 1:45 PM
wendynorris.bsky.social
The post I was responding to (not dunking on) before I was blocked ¯\_ (ツ)_/¯
August 21, 2025 at 12:53 AM
wendynorris.bsky.social
Looks like the de-wokification of the Kennedy Center Honors is a total failure.

KISS' Gene Simmons enthusiastically mused about being a judge on RuPaul's Drag Race.
So, of course we ask Simmons — who to this day still does his own Demon facepaint, because makeup artists “don't know your face, don't know your vibe, don't know your tone” — if he would ever sign up to be a Drag Race guest judge. “Sure!” Simmons says without hesitation. "I’m a big fan of anybody who wants to stand up and say, 'I'm unique, I'm not like everybody else, and you can't judge me. Because the real question is, who the f*** are you?’ … I support LGBTXYZBTA — put all the letters in. I totally support anybody who wants to be unique.”
August 14, 2025 at 2:34 AM
wendynorris.bsky.social
Good morning, scholars
August 1, 2025 at 3:00 PM
wendynorris.bsky.social
So these ChatGPT-souped up librarians invented LibGuides?
July 30, 2025 at 10:50 PM
wendynorris.bsky.social
In a way, it sounds like they are **very belatedly** trying to use the notion of is exclusively licensed content as standing for lawsuits against other LLMs. Good luck with that.

h/t to The Baltimore Banner for this gem though
Currently, the press maintains a backlist of about 3,000 titles and publishes around 150 new books a year. Usually, its authors are faculty and other experts in fields such as public health, science, higher education and the humanities. University President Ron Daniels is one of many authors who received the email this week — he published his book “What Universities Owe Democracy” with the press in 2021.
July 26, 2025 at 2:11 PM
wendynorris.bsky.social
There’s also apparently big interest in building not evil platform alternatives by European social media users.
Not a bad life to ply your engineering and UX wares where life is a little more humane, the food is better, and there’s an appetite for data regulation.
YouGov Poll results that claim more than half of social media users in France, Germany and Spain want non-US alternative platforms
July 23, 2025 at 2:07 PM
wendynorris.bsky.social
I know there are more pressing problems in the world but this will haunt me in my sleep. What does this mean? Ice made with natural water???? What other kinds of water is there, you monsters?
Photo of people shopping at a small streetside market with rows of fresh fruits and vegetables. A sign above the stall reads: "10LB ICE BAG WITH NATURAL WATER $3.49 ea"
July 15, 2025 at 1:49 PM
wendynorris.bsky.social
My name is Kuro T’Chaka. My crunchie bowl is empty. Prepare to die.
July 14, 2025 at 8:19 PM
wendynorris.bsky.social
It's a little more complicated than that. Escalante, the principal who supported his efforts, as well as another teacher left the school due to admin and union opposition.

I was a social worker in the Cleveland schools at the time. A lot of us really looked up to Jaime.

From Wikipedia:
June 29, 2025 at 1:25 AM
wendynorris.bsky.social
The only appropriate response to Academia dot edu violating copyright on my research.
Academia.edu survey form to rate its automated voice over in podcast format of a copyright-protected research paper. Comment: "I did not consent to the use of my work, you thieving bastards."
June 25, 2025 at 4:55 PM