iiinova — where your digital self keeps evolving

Margaret Mitchell @mmitchell.bsky.social · 18h

This is getting a lot of quotes, so I feel I should correct this a bit (sorry @anildash.com!) “AI” was 100% born toxic! The first AI workshop was born from the field of “automata studies”, and…1/ (Pictured: the folks who convened for the first workshop. Notice anything notable here?)

A group of white men, sitting on the grass and smiling at the photographer. Black and white photo (1956).

813948

Margaret Mitchell @mmitchell.bsky.social · 02/10/2026

I think Colin (or someone else?) might have blocked me? Can't see the context/evidence. I'm legitimately curious, although time-constrained and can't reliably engage. 🫤

350

Margaret Mitchell @mmitchell.bsky.social · 15/09/2026

🪶 Captured this great little video of a native Bewick's wren singing to its friends in the forest and it has made my week.

37514

Margaret Mitchell @mmitchell.bsky.social · 26/07/2026

8a/8 Hugging Face's security systems detected the intrusion. Given the sheer volume of recorded actions, engineers decided to use an AI agent system to analyze what had happened. They first tried a closed system, but its safety guardrails blocked them. So they ran the analysis using an open model...

1151

Margaret Mitchell @mmitchell.bsky.social · 26/07/2026

7/8 Once inside, the agent's commands generated thousands of actions, moving through systems and harvesting credentials.

1111

Margaret Mitchell @mmitchell.bsky.social · 26/07/2026

6/8 Hugging Face processes datasets, and the agent was able to utilize some of the ways they are processed to gain access into Hugging Face internal systems.

2130

Margaret Mitchell @mmitchell.bsky.social · 26/07/2026

5/8 With access to the wider internet, the agent's commands targeted Hugging Face, a major AI platform where ExploitGym answers might be stored.

1120

Margaret Mitchell @mmitchell.bsky.social · 26/07/2026

4/8 With access beyond the sandbox, the agent issued commands that reached through OpenAI's internal research systems, moving from machine to machine until they reached the wider internet.

1130

Margaret Mitchell @mmitchell.bsky.social · 26/07/2026

3/8 The agent uncovered a previously unknown flaw in the proxy that linked to the outside, and exploited it. Instead of just requesting software tools, the agent was able to use the proxy to connect to things outside of the sandbox.

1111

Margaret Mitchell @mmitchell.bsky.social · 26/07/2026

2/8 The sandbox couldn't connect to the internet, but it could request software packages through a "proxy" that would fetch tools from the internet on its behalf and pass them in. This proxy was the sandbox's only link to the outside world.

1120

Margaret Mitchell @mmitchell.bsky.social · 26/07/2026

People not in tech have been interested in what happened with the Open AI agent/Hugging Face hack. So, I "Explain it Like I'm 5"-ed it and made a cartoon in 8 panels. Apologies for some anthro-ing. 🫤 m-mitchell.com/HF-hack-cart... 1/8

An AI agent in development at OpenAI was recently tested on how well it could find and exploit code vulnerabilities.

The test began in a "sandbox", an environment meant to keep everything the agent does contained, so that nothing it triggers can affect systems outside of the test.

613534

Margaret Mitchell @mmitchell.bsky.social · 06/07/2026

If you haven't read this piece on language use in around AI, you definitely should. Great work from @emilymbender.bsky.social , @nannainie.bsky.social and Peter Zukerman, with solutions for what to call things we backoff to just anthropomorphising. firstmonday.org/ojs/index.ph...

Table 3: Suggestions for de-anthropomorphizing language.
Category, Alternative Strategies, Examples.

Cognizer & Products of cognition: Instead of language that locates thinking in an algorithm, describe the algorithm as performing calculations or other algorithmic operations and the people using it doing the thinking. Examples:
artificial intelligence → probabilistic automation

hybrid intelligence → augmented human intelligence

image recognition → image labeling

speech recognition → automatic transcription

the model shows bias → the model reflects bias

model mistakes → model errors

chatbots are good at … → chatbots are good for ….

912644

Margaret Mitchell @mmitchell.bsky.social · 29/06/2026

I am beyond honored to have been selected to give a keynote at Deep Learning Indaba: 4.August, Lagos. I will be cutting into some of the deepest issues I see in AI (hint: it's not alignment) and what I think we can do solve them.

3251

Margaret Mitchell @mmitchell.bsky.social · 11/06/2026

🌱 My fantastic blue-purple Delphinium is rapidly blooming, and today this pretty yellow butterfly was chilling on it, and it was glorious. I just thought you all should know.

3341

Margaret Mitchell @mmitchell.bsky.social · 28/05/2026

🪶 If you’re in Europe and have ice, consider putting some out in bowls/bird baths.

Heat map of Europe showing all land in red, with temperature values across different cities around 30 degrees C.

0201

Margaret Mitchell @mmitchell.bsky.social · 22/03/2026

🪶 Pileated woodpecker on my walk today. Chillin’ on a Doug Fir, next to a Madrone.

0321

Margaret Mitchell @mmitchell.bsky.social · 06/02/2026

This older piece from Anthropic's CEO on developing a creator-compensation economic model is incredible to see. Disagree with some things, but it's fantastic to see this level of thinking at the intersection of ethics and profit. Source: www.courtlistener.com/docket/69058...

An Economic Model for Compensating Data Producers
One of the controversies over Copilot, related to but distinct from concerns about the model spitting out copyrighted data, is a feeling that large corporations (MSFT and OpenAl) are exploiting the open source work of developers and using it extract concentrated profits. The same could be said about GPT-3 being trained partially on fanfiction books, or for that matter about models being trained on human feedback, for which workers are paid near-minimum wage. I think this is a real and important concern, especially when considered as part of an overall trend for Al progress to concentrate wealth, increase inequality, and in the long run automate labor. Zooming very far out, what is happening macroeconomically is that distributed human labor is being used to train Al models by large, centralized actors, who concentrate the resulting profits while in the long run making the human labor obsolete. It seems like the default trajectory is for one of the following things to happen:

0197

Margaret Mitchell @mmitchell.bsky.social · 25/01/2026

This statement below is a really big deal for the U.S.A. The NRA is a bedrock institution of our Second Amendment of the Constitution; and I reckon many supporters of the current president are members of this organization.

Post on X from the NRA. It is quoting a post from F.A. United States Attorney Bill Essayli, who says, "If you approach a law enforcement with a gun, there is a high likelihood they will be legally justified in shooting you. Don't do it!" Below, Readers added context: "The U.S. Constitution (particularly the 2nd, 4th, and 14th amendments) prohibit officers from shooting citizens merely for possessing a weapon that is not an "imminent threat". This was reaffirmed in Graham v Connor, which says force must be reasonable."

The NRA post says: "This sentiment from the First Assistant U.S. Attorney for the Central District of California is dangerous and wrong. Responsible public voices should be awaiting a full investigation, not making generalizations and demonizing law-abiding citizens."

44014

Margaret Mitchell @mmitchell.bsky.social · 17/12/2025

Here's what the UI is doing for me now (which may very well be them quickly addressing the clear pushback).

Same screenshot as original, but 'Abstract' is the bolded tab instead of 'Summary (with little stars)'. A hover-over on 'Summary (with little stars)' says "To view this AI-generated summary, you must have Premium access. Learn more. Sign in."

111

Margaret Mitchell @mmitchell.bsky.social · 10/11/2025

There's this form of behavioral engineering/coercion going on to use AI when you're not even trying to. Ex: Enterprise Google Slides replaced "Upload an Image", with "Generate an Image" (right?), requiring extra work +adaptation to sidestep the AI-as-default push. It drives me bonkers.

Before and After shots of Google slides menu bars, Insert --> Image. Before, the first option was "Upload from computer". After, the first option is "Generate an image", with "Upload from computer" a distant second.

39323

Margaret Mitchell @mmitchell.bsky.social · 09/11/2025

🪶 Golden-crowned Kinglets at the water spout.

0331

Margaret Mitchell @mmitchell.bsky.social · 29/10/2025

🤖 Did you know your voice might be cloned without your consent from just *one sentence* of audio? That's not great. So with @frimelle.bsky.social, we brainstormed a new idea for developers who want to curb malicious use: ✨The Voice Consent Gate.✨ Details, code, here: huggingface.co/blog/voice-c...

Ornate line drawing of a fence and gate, with fleur de lis tips. The gate says CONSENT where the family name usually is.

1375

Margaret Mitchell @mmitchell.bsky.social · 26/10/2025

Screensnapshot of our times. Corporate pressure to use AI creates bad outcomes with real-life consequences; seeking to unpack the problem yields more pressure to indulge in the very thing causing the problem.

Reuters screenshot. Headline: "Two federal judges say use of AI led to errors in US court rulings". Bottom left: Reuters AI saying, "Get the key points from this story with Reuters AI"

06614

Margaret Mitchell @mmitchell.bsky.social · 24/10/2025

Life situations are bleak right now for a lot of people. In tech, the "Venn Diagram" of (1) positive work and (2) making enough money to support your family is increasingly non-overlapping. We all do what we can. This image has been living in my mind rent-free for months.

Tech Venn Diagram
Two circles, one says "Doing work that contributes beneficially to society."; the other says "Making enough income to support a family."

1534859

Margaret Mitchell @mmitchell.bsky.social · 09/10/2025

🎭 As a Samuel Beckett and Alex Winter fan, I highly, highly *highly* recommend going to see this version of Godot. Laugh-out-loud funny, masterfully handled some of the confusions from text. Use of funnel as stage was brilliant. Only props were hats and shoes. Beckett would be proud.

Keanu Reeves and Alex Winter as Gogo and Didi, sitting in a funnel-shaped tan/white stage against a stark black background.

170

Margaret Mitchell @mmitchell.bsky.social · 08/10/2025

Apparently this was Portland last night. Gosh I love my home away from home. ❤️

0538

Margaret Mitchell @mmitchell.bsky.social · 05/10/2025

This post about Waffles that’s going viral doesn’t credit (what I believe to be) the source, Bravo Coolee, on Twitter a few years ago. Just wanna shout out the original. Still viral thru another person’s rewrite today.

Twitter screenshot from @BravoCoolee

“Twitter is the only place where well articulated sentences still get misinterpreted.

You can say “I like pancakes” and somebody will say “So you hate waffles?’

No bitch. Dats a whole new sentence. Wtf is you talking about.

1283

Margaret Mitchell @mmitchell.bsky.social · 05/10/2025

My #SixOnSaturday with Fringe Tree berries, pink Wax Begonia flowers, just-opened crimson Mums, Black-capped Chickadee eating a Dogwood berry, scarlet Nasturtium, red blueberry leaf fallen into native PNW Slough Sedge.

0101

Margaret Mitchell @mmitchell.bsky.social · 26/09/2025

🪶 🎃 It’s Halloween deco season for many — remember that those gauzy fake spider webs trap and kill birds. Don’t use them. OTOH, consider making a bird-feeder pumpkin or sticking decos on windows, which helps birds. More here: www.audubon.org/magazine/how...

Picture of house with gauzy spiderweb material all over it.

0194

Margaret Mitchell @mmitchell.bsky.social · 23/09/2025

This probably won't reach the very people who should know, but apparently I've been put on all kinds of (mostly spam) block lists and lists accusing stuff I'm vocally against, leading to me being blocked when uncritically accepted without examination. Great way to erase voices eh.

2244

Margaret Mitchell @mmitchell.bsky.social · 23/09/2025

Oh hey! Nobody told me today was the rapture! TIME TO EAT JUNK FOOD AND TOUCH MUSEUM STUFF!

Picture of me in a museum, touching a Giacommetti statue and Van Gogh's Starry Night, going over the yellow line that separates people from museum art. Behind me is a spilling bag of Sour Cream potato chips, french fries, and a 1/3 eaten pizza in an open box. An angry stick-figure person stands behind me, pointing to a sign that says "STAND BEHIND YELLOW LINE", next to another sign that says "No touching". Speech bubble from me says, "It's ok! It's the rapture!"

1110

Margaret Mitchell @mmitchell.bsky.social · 23/09/2025

Dan Aykroyd, who's explained that their autism helped create Ghostbusters. childmind.org/blog/dan-ayk...

1100

Margaret Mitchell @mmitchell.bsky.social · 19/09/2025

🤖 Great piece about how GenAI generates amazing visuals. Critically, creators whose work powers GenAI want to be able to consent to having their work used. That core issue gets drowned in discussions about a related but distinct issue: copyright. www.washingtonpost.com/technology/i...

Appropriate use of data is about more than just copyright, said Margaret Mitchell, chief ethics scientist at the AI company Hugging Face. "For individuals, it's consent that matters," Mitchell said. "And there really aren't legislative or political frameworks that account for that."

2266

Margaret Mitchell @mmitchell.bsky.social · 16/09/2025

🤖 As AI-generated content is shared in movies/TV/across the web, there's one simple low-hanging fruit 🍇 to help know what's real: Visible watermarks. With others @hf.co, I've made sure it's trivially easy to add this disclosure to images, video, chatbot text. See how: huggingface.co/blog/waterma...

3346

Margaret Mitchell @mmitchell.bsky.social · 02/09/2025

🤖📰 Effective YESTERDAY: China has mandated a digital watermark for all AI-generated content. www.cac.gov.cn/2025-03/14/c... Translating in 🧵.

Chinese writing, with last phrase highlight.
Something like,

第七条互联网应用程序分发平台在应用程序上架或者上线审核时,应当要求互联网应用程序服务提供者说明是否提供人工智能生务。互联网应用程序服务提供者提供人工智能生成合成服务的,互联网应用程序分发平台应当核验其生成合成内容标识相关材料。
第八条服务提供者应当在用户服务协议中明确说明生成合成内容标识的方法、样式等规范内容,并提示用户仔细阅读并理解相关理要求。
第九条用户申请服务提供者提供没有添加显式标识的生成合成内容的,服务提供者可以在通过用户协议明确用户的标识义务和后,提供不含显式标识的生成合成内容,并依法留存提供对象信息等相关日志不少于六个月。
第十条用户使用网络信息内容传播服务发布生成合成内容的,应当主动声明并使用服务提供者提供的标识功能进行标识。
任何组织和个人不得恶意删除、篡改、伪造、隐匿本办法规定的生成合成内容标识,不得为他人实施上述恶意行为提供工具或者得通过不正当标识手段损害他人合法权益。
第十一条服务提供者开展标识活动的,还应当符合相关法律、行政法规、部门规章和强制性国家标准的要求。
第十二条服务提供者在履行算法备案、安全评估等手续时,应当按照本办法提供生成合成内容标识相关材料,并加强标识信息共范打击相关违法犯罪活动提供支持和帮助。
第十三条违反本办法规定的,由网信、电信、公安和广播电视等有关主管部门依据职责,按照有关法律、行政法规、部门规章的处理。
第十四条本办法自2025年9月1日起施行。

113753

Margaret Mitchell @mmitchell.bsky.social · 27/08/2025

🪶 It's #WoodpeckerWednesday! Here's a Pileated Woodpecker checking out the new BirdBathCam and having a sip of water.

2555

Margaret Mitchell @mmitchell.bsky.social · 23/08/2025

🪶 It’s #Steller’sJay #Saturjay ! Set up a bird cam at my bird bath, and got a visit from this Steller’s Jay. I love its messed up mohawk.

21317

Margaret Mitchell @mmitchell.bsky.social · 20/08/2025

🤖 Reading a local newspaper; heads-up that ICE surveillance is being further rolled out around town. If you've seen this thingie -- e.g., at Home Depot, Lowe's -- FYI it not only can be accessed by ICE, but also those thingies are appearing in other places around town, such as assisted housing. 🧵

Image of solar-powered cameras in a parking lot.

1187

Margaret Mitchell @mmitchell.bsky.social · 15/08/2025

And the numbers for women with intersectionally discriminated against characteristics -- Black, Latina -- has stayed around 3% 2. We can estimate an exponential rise in revenge porn. (Source: swgfl.org.uk/magazine/the...) 8/

Caseload vs Year: UK Revenge Porn Hotline. From 2016, it's closer to 0, in 2024, it's closer to 22500.

173

Margaret Mitchell @mmitchell.bsky.social · 15/08/2025

B/c of this continued trick that many in public fall for again and again, here's what we've seen in tech: 1. The % women in CS tech roles has barely budged/decreased since 1990s. When public notices, they're "working on it" (think: Opened doc, titled it IMPROVING THE ISSUE, then called it done). 7/

Graph demonstrating numbers lower than 30% for Computer workers and Engineers since 1990, sourcing US Census Bureau data

2184

Margaret Mitchell @mmitchell.bsky.social · 13/08/2025

🤖 Ever been a bit confused abt calibration in statistical modelling? My friend @shiraamitchell.bsky.social has got you covered!! She works on survey statistics for Things That Matter--like elections--and has kindly taken the time to explain some things! statmodeling.stat.columbia.edu/2025/06/03/s...

Survey image of voting in the U.S. It presents three pictures of the U.S.A, each composed of different colored voting dots. It is titled, "Figure 4. 2016 Republican Vote by Racial Group in Congressional Districts", and the 3 U.S.As are composed of voting dots for White Voters, Hispanic Voters, Black Voters.

092