Mark E. Madsen
@madsenlab.org
Chief Technology Officer at WattIQ
Serial entrepreneur, early at RealNetworks, cofounder Allrecipes, 3 other startups, stint at Microsoft
PhD in Anthropology, studied human evolutionary biology
Board member, Orcas Power and Light Cooperative
Serial entrepreneur, early at RealNetworks, cofounder Allrecipes, 3 other startups, stint at Microsoft
PhD in Anthropology, studied human evolutionary biology
Board member, Orcas Power and Light Cooperative
Reposted by Mark E. Madsen
Sometimes I see people like 20 tweets deep into an argument with Grok. Like, what are you doing brother? You are trying to win an argument with a vending machine.
August 18, 2025 at 3:10 AM
Sometimes I see people like 20 tweets deep into an argument with Grok. Like, what are you doing brother? You are trying to win an argument with a vending machine.
Reposted by Mark E. Madsen
ICE arrested comedian Robby Roadsteamer for singing a parody of Rod Stewart’s “Do Ya Think I’m Sexy” in a Jeffrey Epstein the Giraffe costume outside their facility.
He sings “If you hate brown people / and you are a Nazi” and ICE thugs grab him and drag him into detention. No violence—just speech.
He sings “If you hate brown people / and you are a Nazi” and ICE thugs grab him and drag him into detention. No violence—just speech.
October 15, 2025 at 4:55 PM
ICE arrested comedian Robby Roadsteamer for singing a parody of Rod Stewart’s “Do Ya Think I’m Sexy” in a Jeffrey Epstein the Giraffe costume outside their facility.
He sings “If you hate brown people / and you are a Nazi” and ICE thugs grab him and drag him into detention. No violence—just speech.
He sings “If you hate brown people / and you are a Nazi” and ICE thugs grab him and drag him into detention. No violence—just speech.
Reposted by Mark E. Madsen
This Supreme Court will do anything it can to ensure the end of our representative democracy.
This is a naked effort to save a racist, nihilistic, principle-free far-right minoritarian movement from disappearing from American politics by instead giving it hegemony over the United States... forever.
This is a naked effort to save a racist, nihilistic, principle-free far-right minoritarian movement from disappearing from American politics by instead giving it hegemony over the United States... forever.
If the Supreme Court strikes down Section 2 of the 1965 Voting Rights Act, which has been interpreted to require the creation of majority-minority districts, Republicans could eliminate upward of a dozen Democratic-held districts across the South. Here's how.
The Supreme Court Case That Could Hand the House to Republicans
Democrats would be in danger of losing around a dozen majority-minority districts across the South if the court struck down part of the Voting Rights Act.
nyti.ms
October 15, 2025 at 5:02 PM
This Supreme Court will do anything it can to ensure the end of our representative democracy.
This is a naked effort to save a racist, nihilistic, principle-free far-right minoritarian movement from disappearing from American politics by instead giving it hegemony over the United States... forever.
This is a naked effort to save a racist, nihilistic, principle-free far-right minoritarian movement from disappearing from American politics by instead giving it hegemony over the United States... forever.
Spot on.
Emerging from the darkness of this era will require us to acknowledge that the previous status quo, including many institutions that may have served us well in another era, have been failing for my entire lifetime.
October 15, 2025 at 5:57 PM
Spot on.
Reposted by Mark E. Madsen
you fucked around and now the Episcopalians are doing memes. are you happy now. are you
October 15, 2025 at 4:10 PM
you fucked around and now the Episcopalians are doing memes. are you happy now. are you
@washingtonpost.com - I have resisted the urge to drop my subscription, arguing that we need to support investigative journalism or it won't exist anymore. Your firing of @karenattiah.bsky.social is the last straw. I cancelled today. The WaPo that helped preserve democracy in the 1970's is dead.
September 15, 2025 at 7:34 PM
@washingtonpost.com - I have resisted the urge to drop my subscription, arguing that we need to support investigative journalism or it won't exist anymore. Your firing of @karenattiah.bsky.social is the last straw. I cancelled today. The WaPo that helped preserve democracy in the 1970's is dead.
Reposted by Mark E. Madsen
"Harvard can't use race as a factor in admissions, but ICE can use race as a factor in detentions" is a retrenchment essentially to a pre-Civil War understanding of the Constitution. It's vanishingly few steps removed from "Latinos have no rights which the white man is bound to respect."
September 9, 2025 at 4:44 PM
"Harvard can't use race as a factor in admissions, but ICE can use race as a factor in detentions" is a retrenchment essentially to a pre-Civil War understanding of the Constitution. It's vanishingly few steps removed from "Latinos have no rights which the white man is bound to respect."
Reposted by Mark E. Madsen
The reason I cannot use "AI" in my writing is because I am not speaking merely to be heard to speak.
I am speaking because *I have something to say.*
The things I have to say could not be said by anyone else because these words are mine.
If you don't understand that, I genuinely can't help you.
I am speaking because *I have something to say.*
The things I have to say could not be said by anyone else because these words are mine.
If you don't understand that, I genuinely can't help you.
August 22, 2025 at 5:29 PM
The reason I cannot use "AI" in my writing is because I am not speaking merely to be heard to speak.
I am speaking because *I have something to say.*
The things I have to say could not be said by anyone else because these words are mine.
If you don't understand that, I genuinely can't help you.
I am speaking because *I have something to say.*
The things I have to say could not be said by anyone else because these words are mine.
If you don't understand that, I genuinely can't help you.
“Slight changes to wording that correspond to large changes in meaning” — this, of course, comes from the token embedding which creates a “semantic space”, but one which is not necessarily the same as the semantic space that humans learn in early childhood development.
Bookmarking: “slight changes to wording that correspond to large changes in meaning lead to notable discrepancies between LLMs' and human responses,” even in niche LLMs *designed* as human psych models. arxiv.org/abs/2508.06950
Large Language Models Do Not Simulate Human Psychology
Large Language Models (LLMs),such as ChatGPT, are increasingly used in research, ranging from simple writing assistance to complex data annotation tasks. Recently, some research has suggested that LLM...
arxiv.org
August 14, 2025 at 7:40 PM
“Slight changes to wording that correspond to large changes in meaning” — this, of course, comes from the token embedding which creates a “semantic space”, but one which is not necessarily the same as the semantic space that humans learn in early childhood development.
Shameless repost, because it’s hilarious and true.
August 14, 2025 at 7:33 PM
Shameless repost, because it’s hilarious and true.
Reposted by Mark E. Madsen
Everyone thinks they will be passengers on the Fhloston Paradise, not the janitors. (This is a Fifth Element movie reference).
Let’s see what science fiction has to say about jobs in space…
Let’s see what science fiction has to say about jobs in space…
Made in a lab to make other billionaires seem more palatable.
August 13, 2025 at 2:21 PM
Everyone thinks they will be passengers on the Fhloston Paradise, not the janitors. (This is a Fifth Element movie reference).
Let’s see what science fiction has to say about jobs in space…
Let’s see what science fiction has to say about jobs in space…
Skills, in general, are "use it or lose it" -- at least, if you don't use them, you'll suffer some erosion in your level of performance. I've been saying this for years with mobile mapping and traffic apps -- navigation skills and mental maps are use-it-or-lose-it.
“The AI in the study probably prompted doctors to become over-reliant on its recommendations, ‘leading to clinicians becoming less motivated, less focused, and less responsible when making cognitive decisions without AI assistance,’ the scientists said in the paper.”
AI Eroded Doctors’ Ability to Spot Cancer Within Months in Study
Artificial intelligence, touted for its potential to transform medicine, led to some doctors losing skills after just a few months in a new study.
www.bloomberg.com
August 13, 2025 at 3:03 PM
Skills, in general, are "use it or lose it" -- at least, if you don't use them, you'll suffer some erosion in your level of performance. I've been saying this for years with mobile mapping and traffic apps -- navigation skills and mental maps are use-it-or-lose-it.
Agreed. This is a useful article on LLM hallucinations. Table 3 in particular on root causes is something that everyone working with these models (at the technical level) should look at.
*This is an edifying article about LLM "hallucinations" but I don't think this taxonomy is actually "comprehensive"
arxiv.org/pdf/2508.01781
arxiv.org/pdf/2508.01781
August 13, 2025 at 2:55 PM
Agreed. This is a useful article on LLM hallucinations. Table 3 in particular on root causes is something that everyone working with these models (at the technical level) should look at.
One of the most important AI papers of the year so far. Basically, part of the structure of the internal representation is transmitted to “child” models simply in question output. Fascinating, but highly problematic since it means curing bias is probably no easier among LLMs it is in humans.
Subliminal Learning in AIs
Today's freaky LLM behavior: We study subliminal learning, a surprising phenomenon where language models learn traits from model-generated data that is semantically unrelated to those traits. For example, a "student" model learns to prefer owls when trained on sequences…
Today's freaky LLM behavior: We study subliminal learning, a surprising phenomenon where language models learn traits from model-generated data that is semantically unrelated to those traits. For example, a "student" model learns to prefer owls when trained on sequences…
Subliminal Learning in AIs
Today's freaky LLM behavior: We study subliminal learning, a surprising phenomenon where language models learn traits from model-generated data that is semantically unrelated to those traits. For example, a "student" model learns to prefer owls when trained on sequences of numbers generated by a "teacher" model that prefers owls. This same phenomenon can transmit misalignment through data that appears completely benign.
www.schneier.com
August 4, 2025 at 12:20 AM
One of the most important AI papers of the year so far. Basically, part of the structure of the internal representation is transmitted to “child” models simply in question output. Fascinating, but highly problematic since it means curing bias is probably no easier among LLMs it is in humans.
Good paper with a new benchmark on LLMs and tax prep -- they're not ready for this kind of reasoning yet. Even on simplified sample tests, SOTA models got it right less than a third of the time.
Spreadsheets, calculators, and accountants are still your friends.
Spreadsheets, calculators, and accountants are still your friends.
When my wife used an LLM to do our taxes I thought: we either saved 300 dollars or we're going to jail. Well, now I know which way it's going arxiv.org/html/2507.16...
July 25, 2025 at 5:48 PM
Good paper with a new benchmark on LLMs and tax prep -- they're not ready for this kind of reasoning yet. Even on simplified sample tests, SOTA models got it right less than a third of the time.
Spreadsheets, calculators, and accountants are still your friends.
Spreadsheets, calculators, and accountants are still your friends.
Reposted by Mark E. Madsen
Google's Deep Researcher with Test-Time Diffusion
A new deep research agent designed to mimic the iterative nature of human research, complete with cycles of planning, drafting, and revision.
A new deep research agent designed to mimic the iterative nature of human research, complete with cycles of planning, drafting, and revision.
July 24, 2025 at 1:09 AM
Google's Deep Researcher with Test-Time Diffusion
A new deep research agent designed to mimic the iterative nature of human research, complete with cycles of planning, drafting, and revision.
A new deep research agent designed to mimic the iterative nature of human research, complete with cycles of planning, drafting, and revision.
Reposted by Mark E. Madsen
FWIW, I delivered a paper on this recently at the HyPer workshop at ACM UMAP:
arxiv.org/abs/2505.03434
arxiv.org/abs/2505.03434
Procedural Memory Is Not All You Need: Bridging Cognitive Gaps in LLM-Based Agents
Large Language Models (LLMs) represent a landmark achievement in Artificial Intelligence (AI), demonstrating unprecedented proficiency in procedural tasks such as text generation, code completion, and...
arxiv.org
July 24, 2025 at 6:31 PM
FWIW, I delivered a paper on this recently at the HyPer workshop at ACM UMAP:
arxiv.org/abs/2505.03434
arxiv.org/abs/2505.03434
Reposted by Mark E. Madsen
George Retes, an American citizen disabled US Army veteran who has been held in federal custody since Thursday, was released without charges this evening. 1/
July 14, 2025 at 3:09 AM
George Retes, an American citizen disabled US Army veteran who has been held in federal custody since Thursday, was released without charges this evening. 1/
Reposted by Mark E. Madsen
"NOAA distributed a service change notice to all users, including the National Hurricane Center, that by next Monday, June 30th, they would no longer receive real-time microwave data collected aboard three weather satellites jointly run by NOAA and the U.S. Department of Defense."
Critical Hurricane Forecast Tool Abruptly Terminated
U.S. Department of Defense announced Tuesday it would no longer process and deliver data essential to most hurricane forecasts
michaelrlowry.substack.com
June 26, 2025 at 5:35 PM
"NOAA distributed a service change notice to all users, including the National Hurricane Center, that by next Monday, June 30th, they would no longer receive real-time microwave data collected aboard three weather satellites jointly run by NOAA and the U.S. Department of Defense."
Reposted by Mark E. Madsen
MIT study found that AI overuse causes measurable brain damage. “Cognitive atrophy.. like a muscle that’s forgotten how to work” MIT Publication - arxiv.org/pdf/2506.088...
Time Magazine - time.com/7295195/ai-c...
Time Magazine - time.com/7295195/ai-c...
ChatGPT's Impact On Our Brains According to an MIT Study
The study, from MIT Lab scholars, measured the brain activity of subjects writing SAT essays with and without ChatGPT.
time.com
June 21, 2025 at 1:09 AM
MIT study found that AI overuse causes measurable brain damage. “Cognitive atrophy.. like a muscle that’s forgotten how to work” MIT Publication - arxiv.org/pdf/2506.088...
Time Magazine - time.com/7295195/ai-c...
Time Magazine - time.com/7295195/ai-c...
Apparently, it's been difficult to get Grok to be right-wing enough, just training it on actual knowledge and books and the contents of the Internet.
This is why we need to support the Wikimedia Foundation and other open sources of information and knowledge, more than ever.
This is why we need to support the Wikimedia Foundation and other open sources of information and knowledge, more than ever.
June 21, 2025 at 5:30 PM
Apparently, it's been difficult to get Grok to be right-wing enough, just training it on actual knowledge and books and the contents of the Internet.
This is why we need to support the Wikimedia Foundation and other open sources of information and knowledge, more than ever.
This is why we need to support the Wikimedia Foundation and other open sources of information and knowledge, more than ever.
Reposted by Mark E. Madsen
Remarkable.
MSNC: "I want to reiterate this has been 100% peaceful."
Seconds later, LA County sheriffs start firing on the crowd.
Yet again, "law enforcement" deliberately escalated.
MSNC: "I want to reiterate this has been 100% peaceful."
Seconds later, LA County sheriffs start firing on the crowd.
Yet again, "law enforcement" deliberately escalated.
June 14, 2025 at 11:07 PM
Remarkable.
MSNC: "I want to reiterate this has been 100% peaceful."
Seconds later, LA County sheriffs start firing on the crowd.
Yet again, "law enforcement" deliberately escalated.
MSNC: "I want to reiterate this has been 100% peaceful."
Seconds later, LA County sheriffs start firing on the crowd.
Yet again, "law enforcement" deliberately escalated.
Reposted by Mark E. Madsen
I would like to nominate Maxwell Smart for national security advisor.
Pm Private GIF
ALT: Pm Private GIF
media.tenor.com
March 25, 2025 at 1:32 AM
I would like to nominate Maxwell Smart for national security advisor.
Reposted by Mark E. Madsen
A recent study found that cervical cancer deaths in young women have plummeted since the introduction of the human papillomavirus vaccine.
As Cervical Cancer Deaths Plummet, Experts Credit HPV Vaccine
HPV can cause a variety of cancers, including cervical. New mortality data for women under 25 point to the success of the HPV vaccine.
buff.ly
March 11, 2025 at 12:37 PM
A recent study found that cervical cancer deaths in young women have plummeted since the introduction of the human papillomavirus vaccine.