Timnit Gebru
@timnitgebru.blacksky.app
Personal Account Author: Deep Unlearning: The rise of AI and the radicalization of a tech idealist. Founder: The Distributed AI Research Institute @dairinstitute.bsky.social.
stop. making. people. feel like. chatbots. are "conversational" and their friends.
Mark Follman@markfollman.bsky.social · 23h agoMy new investigation into a mass shooter’s harrowing history with ChatGPT reveals numerous red flags—and some very disturbing ChatGPT replies—from long before his attack on Florida State University www.motherjones.com/media/2026/0... 1x 🧵
In fact, one of the hard things about training & evaluating ML systems, even without an LLM or agent in sight, is making sure the system is learning to solve the real problem, not exploiting quirks of your data set and/or evaluation. There are so many ways to open the door to “cheating”.
I remember a paper from several years ago about classical RL agents learning to exploit bugs/limits in their training gyms to “solve” the problem. One was a little physics sim, the agent learned to exploit the sim’s floating-point error to fly across the world at impossible speed to meet the goal.
Timnit Gebru@timnitgebru.blacksky.app · 5d agoWe're in the era of incompetence & cybercrimes headlined as "unprecedented model capabilities gone rogue." So OpenAI & Anthropic are trying to one up each other with such incompetence because the "press release as a service" performing media & clueless politicians parrot pre IPO CEO talking points.
My car went rogue after I turned on cruise control and jumped out
Timnit Gebru@timnitgebru.blacksky.app · 5d agoWe're in the era of incompetence & cybercrimes headlined as "unprecedented model capabilities gone rogue." So OpenAI & Anthropic are trying to one up each other with such incompetence because the "press release as a service" performing media & clueless politicians parrot pre IPO CEO talking points.
We're in the era of incompetence & cybercrimes headlined as "unprecedented model capabilities gone rogue." So OpenAI & Anthropic are trying to one up each other with such incompetence because the "press release as a service" performing media & clueless politicians parrot pre IPO CEO talking points.
"As Japanese financial newspaper Nikkei Asia found in a recent investigation, just five US tech giants — Alphabet, Microsoft, Amazon, Meta, and Oracle — are hiding an estimated $1.65 trillion in debt that doesn’t appear on balance sheets. futurism.com/artificial-i...
AI Companies Are Trying to Hide a Staggering Amount of Debt
AI companies are pouring tens of billions of dollars into enormous data centers. They're being built on top of a mountain of hidden debt.
futurism.com
Hard agree. We’ve seen so many examples of people who consider themselves ‘progressive’ try to weaponize their belief systems to specifically target Black and Brown folks and dismiss obvious and chronic issues we face. In this context, they are no different from a common white supremacist
Timnit Gebru@timnitgebru.blacksky.app · 7d agoJust wanna say that I don't care if you consider yourself "left." If you're a tankle supporting genocides around the world that don't violate your narrow conception of harm, gaslighting victims, and coming here to demind my time & paternalistically engage with your colonial self, I'm blocking you.
Just wanna say that I don't care if you consider yourself "left." If you're a tankle supporting genocides around the world that don't violate your narrow conception of harm, gaslighting victims, and coming here to demind my time & paternalistically engage with your colonial self, I'm blocking you.
Let me just say FUCK YOU to all the people who believe Africans should just shut up and put up with the "right" voices that gaslight and support our genocides. Fuck you.
I say someone is genocidal, I give specific examples of how they are genocidal towards my own people and communities, and someone on here says "but they've spoken up about Palestinians, what are your sources for your claims?"
Look at Democracy Now reposting this platforming of BreakThrough News to hide critical comments about the genocidal tankie news organizations they're platforming. Because millions of communities around the world harmed by them should be quiet since they're "left" in the western left/right dichotomy.
Democracy Now!@democracynow.org · last wk.House Republicans have subpoenaed the independent news outlet BreakThrough News, demanding its internal communications and financial records. "This is a flagrant attack on our First Amendment rights as an independent media outlet, and everybody should be alarmed," says @raniakhalek.bsky.social.
Rania Khalek has been an antiblack piece of shit for well pver a decade now, astounding. We had to gather her back in 2014 around her denial of antiblackness and she's STILL at it, SMDH
Timnit Gebru@timnitgebru.blacksky.app · last wk.You are disgusting for platforming someone as genocidal as Rania Khalek and the so-called BreakThrough "news" organization. These are the victims she went out of her way to gaslight. These are the perpetrators she said were falsely accused of genocide. www.theguardian.com/global-devel...
Democracy Now, a news outlet I have been following for YEARs, platforming someone as genocidal as Rania Khalek, is disgusting.
Democracy Now!@democracynow.org · last wk.“Everybody who cares about a free press and cares about free speech should be alarmed about an attack like this," says BreakThrough News journalist Rania Khalek.
I'm going to choose to believe from now on that singularity is the ultimate form of the incel
Futurism@futurism.com · last wk."I've been waiting for this my whole life."
Conversely if everyone "in tech" knows this is kayfabe (don't break the kayfabe), it's a cynical move by an entire professional field to accrue even more power to themselves using participatory make-believe against the informed consent of the rest of the world at large.
If everything OpenAI and Anthropic are saying about their own products is true, it is the height of irresponsibility and unsafety that anyone is intentionally choosing to use them.
It's also a problem that this inverted approach to agency has already trickled out to the broader developer community and tech industry, who do things like complain that literally having to lift a finger to give permission to the agents to do whatever is too much work. The drag of being in the loop.
They are intentionally cosplaying the precusor events for an apparent Neuromancer-style AI breakout scenario, relying on social knowledge of this science fiction idea existing in the public noosphere and the credulity of the tech press and hype of tech workers to carry the collective belief home.
The role of agency and the concept of the agent plays a disturbing double motion where the creators of the systems role-play the pretense that they don’t have agency over their own design decisions in order to create a self fulfilling prophecy of “the runaway agent”.
What disturbs me about the practices and myth-making of the AI labs is they are intentionally using bad and unsafe design to create systems that they make-believe are out of their control in order to use this as a forcing function to get governments and civil society to cede them more power
Timnit Gebru@timnitgebru.blacksky.app · last wk.I never called things that happened in my circuits "rogue systems." I called it "bad design on my part." But here we are.
WELP I WAS ANNOYED AND HAD THE TIME: “When men say she should be grateful for the attention, they’re saying a woman’s scholarship only gains true meaning when a famous man shines a light on it.” news.chanda.science/archive/an-o...
An Odyssey Through Testimonial Injustice
On reading Dr. Emily Wilson, properly
news.chanda.science
Chanda Prescod-Weinstein 🌌@chanda.blacksky.app · last wk.Don’t make me write a newsletter about how Emily Wilson is allowed to be an expert, you guys
The Hugging Face incident was “both notable and alarming,” EFF’s @evacide.bsky.social told @sfexaminer.bsky.social, “not because ‘ooh, the model’s so powerful,’ but because it demonstrates a colossal failure on the part of OpenAI to secure the sandbox in which it was testing its model.”
OpenAI hacking incident highlights AI cyberrisks
Experts take company to task for safety efforts
sfexaminer.com
It's all a bunch of closely related computational techniques, and the name is whatever lands external money for your efforts at that point in time.
Timnit Gebru@timnitgebru.blacksky.app · last wk.Now is a good reminder that machine learning is a rebranding of statistical learning. Reinforcement learning is something I first came across in control theory. Many ML formulations are from information theory, a much more theoretically grounded discipline that's been around for decades.
When I’m called on to explain LLMs, I almost always start with “so once upon a time thereabouts in the era of WW2, there was a guy called Claude Shannon…” who is far more fun to talk about than sam altman or whatever, but it also guarantees whoever I’m talking to will never ask me about LLMs again
Timnit Gebru@timnitgebru.blacksky.app · last wk.Now is a good reminder that machine learning is a rebranding of statistical learning. Reinforcement learning is something I first came across in control theory. Many ML formulations are from information theory, a much more theoretically grounded discipline that's been around for decades.
Repeatedly pointing at the sign as I go rapidly insane in this accursed industry
Timnit Gebru@timnitgebru.blacksky.app · last wk.Tech reporters not only do press-release-as-a-service, but also want to regularly talk to those of us who have been working in these disciplines from so many angles, with their uninformed and credulous takes🙄
