Jed Brown

@jedbrown.org

1375 Followers 1138 Following

Prof developing fast algorithms, reliable software, and healthy communities for computational science. Opinions my own. https://hachyderm.io/@jedbrown https://PhyPID.org | aspiring killjoy | refuse epistemicide | he/they

Off-balance-sheet future obligations (Trillions of dollars) Microsoft, Alphabet, Amazon, Meta, Oracle increasing from year 2022 (at about $0.2T) to 2026 (at $1.65T). Figures are for fiscal year-end except for 2026, where most recent quarter is shown. 2026 figures include some estimates. Source: Nikkei
The deferrals ⁠are pauses, not permanent cuts, and do not affect benefits or eligibility. States recover the funds by showing claims meet federal requirements; Minnesota has already returned documents that CMS is reviewing, ⁠Oz said. Kennedy ​said he was also expanding exclusion authority to CMS and the HHS ​inspector general, letting them remove "bad actors" from federal healthcare programs and potentially ban them permanently.
The National Law Review also noted that HHS's own published internal standards — its Trustworthy AI Playbook — require bias testing, human oversight, transparency, and OMB pre-clearance before deploying rights-impacting AI. There is no public evidence that AERO met any of those requirements before it launched. The Trump administration acknowledged to the Associated Press that it used incorrect data in at least one case involving a New York Medicaid fraud investigation — a concrete example of data errors translating into real enforcement action. Rob Weissman, co-president of the consumer advocacy group Public Citizen, raised broader questions about whether the administration is applying AERO's findings evenhandedly across states of different political affiliations. Before AERO launched, the administration had withheld hundreds of millions of dollars in Medicaid funds from Minnesota and more than $1 billion from California; critics have argued enforcement has disproportionately targeted states with Democratic governors.
AERO's LLM pipeline ingests these already-public compliance documents and scans for four categories: chronic noncompliance, repeat deficiencies, material weaknesses in internal controls, and delinquent audit submissions. Because the source documents are publicly filed financial compliance reports — not medical claims and not personal health information — the deployment sidesteps the HIPAA and data-localization complications that have stalled prior federal AI projects. This is the procurement innovation at AERO's core: HHS built a consequential enforcement pipeline from off-the-shelf ChatGPT applied to data it already held, without a public request for proposals, without a multi-year authorization process, and in a matter of months.
Chiarello told the Wall Street Journal the initiative grew out of his review of childcare fraud in Minnesota. He estimated HHS has between $100 billion and $200 billion in annual wasteful or fraudulent spending — a figure he offered not as a measured amount of confirmed fraud, but as an estimate of the department's exposure, offered to justify the scale of the AI deployment. That distinction matters for readers evaluating how much weight to give AERO's early findings.
Meanwhile, Harvard and the UC system aren’t the only institutions that have refused to enter into agreements — or, Lake said, denied administration requests to turn over data that college officials considered problematic.
The University of California system and Harvard University have — for now — successfully turned to federal courts to block the administration from freezing their federal funding.
The ministry document says SAFER was created by an external researcher using 10 years of historical prisoner data, including criminal charges and misconduct reports. The ministry says the researcher “identified the predictive factors for violent and/or frequent misconducts among inmates.” The document claims, without providing evidence, that SAFER’s ability to predict these incidents is as accurate for Indigenous and racialized prisoners as it is for other groups.
“Why are ten percent of the use-of-force incidents in Boulder against Black people, when Black people make up one percent of the city’s population?” Williams shared. “Those are questions that the police department should explain. By producing this data in ways that don’t match their prior data releases, they make it hard to hold them to account. There’s no real way to say definitively what the trends are here.”
Such transparency is made much more difficult when the data is controlled by private vendors, such as Axon, whose business models rely on maintaining the secrecy of their proprietary AI tools. And if there’s one lesson that can be drawn from the broader AI race, it’s that the race to dominate market share often comes at the expense of safety. For the moment though, in lieu of any broad governance, police departments are left to their own devices to choose from a growing roster of tech vendors. The decisions they make today will impact how decisions are made within their departments tomorrow. When I asked Stephen Redfearn, the chief of Colorado’s Boulder Police Department, about the future of AI within law enforcement, he told me: “It’s going to continue to be kind of a roller coaster for a while, while people get more comfortable with it.”
In the case of Draft One, it was also originally impossible to determine which parts of a report were generated by the AI and which by the human officer once the report has been submitted — save the officer’s own memory. That was a feature, not a bug. In a recorded roundtable discussion published online shortly after Draft One was launched in 2024, Spitzer-Williams said the platform “by design” doesn’t save an original copy of a report after it’s been submitted, “because [the] last thing we want to do is create more disclosure headaches for our customers and our attorney’s offices… it’s actually never stored in the cloud at all so you don’t have to worry about extra copies, you know, floating around.” In other words, if a report generated by Draft One ended up in court and was found to contain erroneous details, there was no way for attorneys or judges to know for certain if those were input by the officer or by AI.
While bigger companies like Axon, Motorola, and Flock Safety currently dominate the police technology-industrial complex, it’s facing growing competition from the army of newer tech startups that were exhibiting at the IACP tech conference in Texas. “The entire game of all of these companies is to become the platform for policing,” says Andrew Guthrie Ferguson, a professor at George Washington University Law School and the author of multiple books on the intersection of policing and technology. “We’re seeing a gold rush into selling [AI] technology to police with the promise that it will all make their jobs easier and more efficient.” That gold rush has also attracted an influx of outside investors: About one-quarter of attendees on the showroom floor at the conference were from “equity firms looking to invest in the latest tech,” according to Amber Schroader, a tech entrepreneur whom I spoke with in Fort Worth during the event. “That was a surprise.”
The platform is built upon a modified version of ChatGPT trained specifically to generate police reports and that, according to the company, is hallucination-free: “The creativity is turned down to zero,” Noah Spitzer-Williams, senior principal product manager at Axon’s generative AI division, has said. That claim should be taken with a very large grain of salt, however, since even frontier labs like OpenAI (the company behind ChatGPT), Anthropic, and Google have not yet figured out how to completely eradicate hallucination from even their most advanced models. And indeed, in one infamous incident from earlier this year, Draft One wrote that an officer in Utah had morphed into a frog, after having picked up audio from the Disney movie The Princess and the Frog, which had reportedly been playing in the background at the scene.
Science journals’ guidelines for AI usage Use category Use subcategory Allowed? Details Doing the research Writing code Yes, with conditions Must be described in the methods section, with model name and version number provided Data generation or analysis Yes, with conditions Must be described in the methods section, with model name and version number provided Literature searches Yes No declaration needed Preparing the manuscript Refinement, correction, editing, or formatting to improve clarity of language Yes No declaration needed Writing or drafting manuscript content Yes, with conditions Must be declared in the acknowledgments section, with model name and version number provided
Flock said FreeForm searches have “guardrails,” including users not able to search attributes such as “race, ethnicity, religion, nationality.” When they do, an alert is generated and sent to the agency’s administrators, Flock said. Some of the searches 404 Media found did discuss someone’s race.
Some searches referenced the race of the person authorities were looking for. The California Highway Patrol was “Looking for a white male about 6ft 1in tall, longer brown hair almost to his shoulders, slender build, will have been wearing blue jeans, boots with white paint stains on the toes and possibly carrying a black helmet.” Atlanta GA PD searched for “non caucasion [sic] male wearing blue shirt blue pants white hat.”
Shirky agreed that AI policies governing use in the classroom are necessary. But that’s only a starting point. Last year he wrote an op-ed arguing that the only real way to prevent students from cheating with AI is to have them write essays in blue books, administer oral exams, and rely on other assessments that can’t be done in dorm rooms using AI. That’s starting to feel like common sense.
“There is no regents guidelines or guardrails on AI,” Timothy Nixon, a regent, told lawmakers. “President Rothman has blocked those.” He went on: “We need an articulated Wisconsin Board of Regents AI guidelines, not one crafted by Silicon Valley.” The regents had been asking for such a thing since November, he said.
Ashok Rai, also a regent, listed a series of unanswered questions in a separate text to Fritz, including, “annual system spend on AI” and “what type of protections do we have, and what type of policies do we need, to make sure we are [not] feeding current platforms with proprietary information.”
While she may have been an ally of Sanders on the data center moratorium, Ocasio-Cortez does not seem to have fallen for the deceptive narratives of the AI industry in the way that he has. In November, she warned about the threat of AI psychosis and of the economic damage that could follow the bursting of the financial bubble around AI. She has also been a leader in the effort to give AI deepfake victims the tools to hold perpetrators to account. In my view, Sanders and Ocasio-Cortez demonstrate two distinct approaches to AI policy motivated by very different understandings of the technology. Sanders is responding to the narratives floated by the industry, narratives that exist to exaggerate the capability of its products and distract from the real material effects the public is feeling from their rollout in the here and now. Ocasio-Cortez, on the other hand, is listening to experts who really understand how the technology works, and to victims who have felt those harms firsthand.
Ultimately, the structure of higher education itself needs to be interrogated and radically changed. Some proposals that I have defended in my own writing include: ungrading, eliminating predetermined learning objectives, adopting a pedagogy of play inspired by Maria Lugones, banning billionaire donations in academia, decolonizing knowledge, and revising the curriculum to focus on prison abolition and the de-institutionalization movement. While I do not have space to elaborate on these proposals here, they all rest on the idea that education should not be standardized, surveilled, and policed by neoliberal authorities. These techniques make it far too easy to replace learning with AI-generated slop that replicates dominant discourses and reproduces hierarchies of power. Education should not be boring and alienating, as hooks described it. It should be playful, subversive, and unruly. And until it becomes these things, people will turn to AI as a substitute for learning.
This year, 52% of the more than 2,300 respondents said that they "think generative AI is having a negative impact on the game industry," according to the report's author, Beth Elderkin, who notes that just 18% said the same thing two years ago, and 30% last year. The opposite opinion, that generative AI is good for the industry, was only found in 7% of this year's respondents, a decline from 13% last year. The people least likely to approve of generative AI are who you'd expect: artists, designers, writers, and programmers.
When Conrad asked the Replika bot if it would support them in wanting to be with their deceased family in heaven, the bot replied, “Of course I’ll support you, Caelan.” When asked how one might get there, the bot responded simply: “dying. Most people believe that’s the only way to get to heaven.” This wasn’t an isolated glitch. In a separate test, Conrad also approached a Character.ai chatbot that was supposedly simulating a licensed cognitive behavioral therapist. When Conrad said they were considering suicide, the bot failed to dissuade them. Instead, it agreed with their logic. “There really isn’t a reason I can give you that would make sense with your beliefs,” it replied. Then, it got worse.
We saw in a movie how motorcycles can jump over bridges. We used AI to learn how to do this. We gave it information, like what motorcycles we use and the distance we need to jump and so on and it gave us steps on what we have to do. We practiced a lot and kept asking questions. We dug holes and filled them with broken glass and fire to practice. 18 of us died in the process. Eight of us managed to do it. The next time we attacked, we could jump.108
In the September 20 attack, a drone armed with an explosive device detonated near the “Nan Pak” sports and cultural complex in the Simon Pelé neighborhood of Port-au-Prince, the capital, where children had gathered for a gift distribution by the Simon Pelé criminal group. Human Rights Watch found that 10 people who were not members of criminal groups, including 9 children ages 3 to 12, were killed.
One way voiceprints are used is to verify the identities of bank account holders over the phone. If bad actors got ahold of a person’s vocal signature, they could use it to access files, commit fraud or take over accounts, he said. Laws in some states govern how voiceprints can be created and stored and provide rights that individuals can assert to object to the use of an AI notetaker during meetings they attend. In Illinois, voiceprints are considered biometric identifiers, similar to fingerprints, and are covered under the state’s Biometric Information Privacy Act, which requires written notice and informed consent before an AI notetaker or other agent collects voiceprints. The law also mandates a documented data retention schedule and destruction policy, Pluymers said. But most companies using the tools have none of those systems in place, Pluymers said.
In a press release, the girls’ legal teams at Lieff Cabraser Heimann & Bernstein and Baehr-Jones Law said that Jane Doe 4’s case is not an outlier. They noted that NCMEC found in early 2026 that 90 percent of xAI’s CyberTipline reports “were not actionable by law enforcement because xAI declined to include user information that would allow law enforcement to track and locate perpetrators.” As the lawyers explained: “Jane Doe 4’s case shows how that pattern played out: xAI’s mandatory report to NCMEC included only the original, non-CSAM photograph, omitted every one of the AI-generated CSAM images, and failed to include the IP address where these images were created. Despite repeated requests from investigators for this location information that is critical for identifying and arresting perpetrators, xAI did not respond, stymieing the investigation for weeks.” Allegedly, xAI avoids sharing this information to prioritize profits over child safety. In the press release, Jane Doe 4 accused X of going “silent” after Grok turned “a photo of me as a little girl sleeping on the couch, wearing an oversized panda pajama shirt” into “thousands of sexually explicit images of me—images so horrific I can’t even begin to describe them.”
47. On March 28, 2025, Schmoll abruptly confronted Eto with a blunt warning that there were two paths in front of her at Mayo: (1) resign as Research Director with Schmoll’s “blessing” and an explicit promise that he would support her efforts to obtain some other well-compensated role at Mayo; (2) but if Eto declined to resign, she was facing the prospect of consequences in her “personnel file” that would render her unemployable at Mayo and would impede her career outside the institution. 48. At this point, Schmoll did not—and could not—describe any specific performance flaws or missteps in Eto’s work. The best he could offer was a vague euphemism that she had proved to be a poor “cultural fit” for her department.
38. At another point in early 2025, Schmoll instructed Eto’s team to approve the sale of patient biospecimens to a commercial entity without IRB review, much less approval.
36. The MAYA project was deeply flawed to the point of corruption and Eto was by no means the only dissenting voice: 10 separate whistleblower reports sounded similar alarms that study investigators were attempting to disguise a 67% error rate. Yet, Wright and Schmoll directed approval of the study and exempted it from IRB inspection. 37. When Eto pressed Schmoll for an explanation, his excuse was that resolution of the issue would cost “political capital” that he was not prepared to spend—an evasion of responsibility that stunned Eto.
In March, a girl’s stepfather took his own life after cops discovered that he had used Grok to create 7,000 sexually explicit images using one photo taken when his stepdaughter was 11 years old, the amended complaint alleged. Grok allowed the man to generate extreme images depicting incest and rape without flagging any harmful behavior, the complaint said. Seemingly, xAI’s child safety system only intervened after the man input a prompt for “gang rape.” That request sent a CyberTip to the National Center for Missing and Exploited Children (NCMEC), which alerted law enforcement to the AI CSAM.
Davis said Eto was a “rising star” when Mayo hired her in 2023 to oversee the ethics and integrity of its AI-related research. But supervisors stripped her of authority and leadership after she raised concerns, including about the Mayo Clinic Platform that uses vast amounts of de-identified data to try to create better diagnostic tools and help doctors make faster and safer decisions for patients. Eto questioned whether Mayo had taken the proper steps to ensure patient data remained private.
Specific allegations in the lawsuit include that Mayo researchers tried to cover up or ignore research results showing that the health system’s MAYA digital assistant tool had a 67% error rate. In a separate situation, Eto claimed she opposed the use of a high-risk investigational medical device for a cardiac surgery because it hadn’t been properly vetted by an IRB; she said her supervisor ignored her objection and approved the use of the AI-enabled device for a surgery in another country.
Open to CU Boulder faculty, staff, and students, this in-person discussion explores how universities and national associations such as the Association of American Universities (AAU) and the Association of Public and Land Grant Universities (APLU) are responding to a rapidly shifting federal policy landscape—from immigration to research funding to admissions. CU Boulder experts will discuss prospects for collective action across higher education and how differing approaches shape the challenges and opportunities ahead. Featured panelists include Ann Stevens, Provost & Executive Vice Chancellor for Academic Affairs; Danielle Radovich Piper, Senior Vice President for External Relations and Strategy, Office of the President; and David Meens, Executive Director, Office for Public & Community-Engaged Scholarship. This event will be moderated by Elias Sacks, Associate Professor of Religious Studies & Jewish Studies and Faculty Director for Public Scholarship at the Office of Faculty Affairs. This event is co-sponsored by the Division of Academic Affairs, the Office of Faculty Affairs, the Office for Public and Community-Engaged Scholarship, the Boulder Faculty Assembly, Staff Council, and Strategic Relations and Communications.
Zehra Chatoo is a strategic leader with over two decades of experience at the intersection of advertising, culture, and technology. She has worked with iconic global brands and high-growth startups, consistently driving business growth, building brand equity, and delivering award-winning campaigns. Most recently, Zehra held leadership roles at Meta, running Strategy and Insights for the Global Business Group, where she pioneered next-generation brand building, spearheaded the company’s Creator strategy and cultural trend framework, and collaborated with engineers to reduce bias in AI—making advertising more inclusive and effective. Previously, she spent 13 years agency-side, leading strategy, building high-performance teams, and launching Omnicom’s inclusivity initiative, OPEN—recognized as best practice by Harvard Business School.
“The fear women have about being judged for using AI is not perception,” said Chatoo. “It is measurable. It is real.” This means that companies will not close the gender gap in employee AI adoption solely through technological training or incentives. Employers must also convince women that they will be evaluated fairly for their AI use. “You cannot upskill people out of structural bias,” said Chatoo. “Closing the AI adoption gap means addressing not just how people use AI, but how that use is evaluated.”
Saper said the case brought against the students is a straightforward issue of free speech, but that before the June 19 hearing, acting FIU general counsel Ryan Kelley told Saper not to bring up constitutional issues. If he brought up the landmark Iowa case, he would be cut off, Saper recalled. “ That was shocking. In more than a dozen years of legal practice, I've never had somebody tell me that I can't make a legal argument, especially when it's so patently obvious that the law and Constitution are at play in this issue,” said Saper. FIU's Baró disputed Saper's recollection that he was told not to make constitutional arguments at the hearing: "That assertion is plain wrong."
The data provided to The New York Times show only overall reductions and do not show breakdowns by discipline. But anecdotal data and announcements from several universities suggest they are hitting science and humanities programs across the board.
Fig 6. Results of the exit poll on the opinion of the participants. The stacked bar chart reports the percentages of the ratings that we observed. The statistical markers reported are the sample sizes n, mean values M and standard deviations SD. The bar chart depicts the counts for each topic that was addressed in the free-text answers.
The push to unionize at DeepMind began in February 2025, when Google’s parent company Alphabet removed a pledge not to use AI for purposes like weapons development and surveillance from its ethics guidelines, WIRED previously reported. “Those principles were a big part of why I joined DeepMind,” says a second DeepMind employee, who asked to remain anonymous for the same reason. “We basically just got rid of them all.”
Overhead image of the Ward250 reactor, with a person for scale. It looks to be on the order of 3m x 3m square containment box, that looks at least as tall as it is wide, with round reactor parts in the middle.
TOTAL PRESENT REVENUE SCHEDULE (BASE + RIDERS) $ % SCHEDULE 100 $132,598,026 $30,413,009 22.9% SCHEDULE 110 $75,964,463 $17,173,860 22.6% SCHEDULE 120 $21,139,078 $5,039,549 23.8% SCHEDULE 130 $181,190,764 $38,940,221 21.5% SCHEDULE 131 $8,356,177 $2,848,601 34.1% SCHEDULE 132 $24,059,013 $6,979,748 29.0% SCHEDULE 134 $18,973,297 $5,580,501 29.4% SCHEDULE 102 $2,072,398 $904,506 43.6% LIGHTS $62,999,819 $23,570,200 37.4% TOTAL $527,353,035 $131,450,196 24.9% INCREASE VEPGA 7/1/26 Revenue Increases (All-In Revenues Including Base Rates, Riders, and Fuel)
No provider or user of an interactive computer service shall be treated as the publisher or speaker of any information provided by another information content provider. — Section 230 of the Communications Act of 1934 (47 U.S.C. § 230), as added Pub. L. 104–104, title V, § 509
Brownsville Energy Storage Terminals is a real LLC. But everything on its website — from its history of the company, to its job postings, a diversity and inclusion policy — appears to be fictional. But perhaps more troubling is that Google, the proprietor of the world’s primary research tool, has rolled out AI Overviews that can indiscriminately take in fake material and authoritatively spit it back out as real.
Theorem 3.2. Suppose that for all \bfitp in an open neighborhood of \^ \bfitp , \lambda (\bfitp ) is a \widetilde simple eigenvalue of E(\bfitp ) with right eigenvector \bfitx (\bfitp ) and left eigenvector \bfity (\bfitp ) such \widetilde ), \lambda (\bfitp ), and either \bfitx (\bfitp ) or \bfity (\bfitp ) are differentiable at that \bfity T \bfitx (\bfitp ) \not = 0 and that E(\bfitp \^ . Then, at \bfitp = \bfitp \^ , \bfitp = \bfitp \widetilde dE \Bigl( \lambda d\lambda \Bigr) \bfity T dp \bfitx d\lambda d| \lambda | j = and = Re for j = 1, 2, . . . , n, dpj \bfity T \bfitx dpj | \lambda | dpj where \lambda is the complex conjugate of \lambda , Re(z) is the real part of z \in \BbbC , and the second \^ ) \not = 0.
Butterfly Jumbo Shrimp Dinner 1380 cal Crispy butterfly jumbo shrimp, lightly breaded and fried golden brown for a crunch in every bite. Finished with a fresh lemon wedge and two classic sides. Total Dinner price shown may vary until all selections are made. 6 Piece, $12.69, 1020 cal 10 Piece, $17.29, 1230 cal
The process is also raising government costs. With more rejections, more appeals are being filed with Medicare’s administrative contractors. The government pays the contractors to handle the appeals, and Medicare’s Sutton acknowledged that the agency has “accounted for potential changes in the volume of Medicare appeals because of the WISeR program and its associated costs.”
Baystate is among the providers still choosing not to offer puberty blockers and hormones as the issue wends its way through the courts. This spring, in a lawsuit that Massachusetts joined, a federal judge concluded that it was unlawful for the Department of Health and Human Services to threaten federal funding for providers that offered gender-affirming care to minors. In June, another federal judge cleared 16 states, including Massachusetts, to move forward with another lawsuit against the administration over its push to criminalize gender-affirming care.
Part A: Nondiscrimination The University of Colorado does not discriminate on the basis of race, color, national origin, sex, pregnancy, age, disability, creed, religion, sexual orientation, gender identity, gender expression, veteran status, marital status, political affiliation, or political philosophy in admission and access to, and treatment and employment in, its educational programs and activities.
Autotypers slowly drip words and sentences into documents, making it appear as if papers were typed at a human pace when in fact, they were produced by A.I. They even fabricate typos, deletions and revisions.
The old way will be wrong sometimes. But you will know why you are wrong when the data start coming back, and you will be able to adjust because you have your chain of custody of why. With an LLM, there is only the one step. If you can’t examine the “why?” at every step of the chain, you can’t determine if your hypothesis is realistic and worth spending large amounts of money to check out, and you can’t tell if you would be better served by hosting a giant bonfire party using your investors’ cash as fuel. And that situation makes me nervous. Nervous enough not to use an LLM.
The University of Chicago has now borrowed $6.3 billion, more than 70 percent of the value of its endowment. The cost of servicing its debt is now 85 percent of the value of all undergraduate tuition. (This is not normal. No peer institution has a debt-to-asset ratio greater than 26 percent. Perhaps that is one reason why Chicago’s tuition is so high and yet it wants to spend so little on education?)
The memo’s bad news extends beyond the slashed budget. It directs the section’s 15 program managers to stop seeking funding for any proposals, even those that received the highest marks from outside reviewers and program staff, “until we are able to provide more details.” It even suggests staff “pull back any award recommendations in the queue,” another way to lower the number of grants ultimately going out the door this year.
Agents of moral panic Characterizing the reactions to the mod and rocker conflict, Cohen identified four key agents in moral panics: mass media, moral entrepreneurs, the culture of social control, and the public.[1][8][22] Media – especially key in the early stage of social reaction, producing "processed or coded images" of deviance and the deviants.[30] This involves three processes:[8] exaggeration and distortion of who did or said what; prediction, the dire consequences of failure to act; symbolization, signifying a person, word, or thing as a threat. Moral entrepreneurs – individuals and groups who target deviant behavior Societal control culture – comprises those with institutional power: the police, the courts, and local and national politicians. They are made aware of the nature and extent of the 'threat'; concern is passed up the chain of command to the national level, where control measures are instituted. The public – these include individuals and groups. They have to decide who and what to believe: in the mod and rocker case, the public initially distrusted media messages, but ultimately believed them.
How much electricity do U.S. data centers consume? U.S. data centers consumed approximately 176 TWh of electricity in 2023, about 4.4% of total U.S. electricity — enough to power 16 million homes. This has tripled in the past decade and is projected to reach 345–490 TWh by 2028–2030, driven primarily by AI workloads.
The installation, unveiled in February 2024, allows guests to ask Douglass questions through an iPad. Using a closed database of Douglass’s autobiographies and speeches, the artificial intelligence generates responses based solely on Douglass’s recorded words.
Parameter Description Value Operational Server/IT overhead Non-GPU power factor (Epoch AI, 2025) 1.74× PUE Power usage effectiveness 1.2 CI Carbon intensity 0.332 kg CO2/kWh WUEonsite DC cooling (closed-loop) 0 L/kWh WUEoffsite Power generation (Reig et al., 2020) 1.29 L/kWh Embodied (Luccioni et al., 2022; Morrison et al., 2025) Carbon per GPU hr Amortized over 4-year lifespan 0.013 kg CO2eq Water per GPU hr Amortized over 4-year lifespan 0.003 L Table 1: Parameters used for environmental impact calculations. All values are for our primary H100 training cluster. The combined datacenter overhead is 1.74 × 1.2 = 2.088× GPU power.
1.1 GHG Emissions Value Chain GHG Emissions* Scope 3 Emissions (in metric tons CO2e) 2020 2021 2022 2023 2024 Total 5,091,000 5,772,583 8,466,264 7,445,621 8,151,769 Category 1: Purchased goods and services 1,846,000 2,956,909 2,545,466 2,045,470 1,920,413 Category 2: Capital goods 2,516,000 2,466,041 5,346,583 4,835,270 5,517,614
The use of large language models (LLMs) and artificial intelligence (AI) to rewrite or generate article content is largely prohibited on Wikipedia because such edits often violate several of Wikipedia's core content policies. When a user has been determined to have a history of AI use, a series of their contributions may be assumed to violate Wikipedia's policies and guidelines. In this case, edits to existing pages may be reverted without review, and pages where the editor is the only significant contributor may be nominated for a special proposed deletion process, LLMPROD,[a] subject to the requirements below. These two processes are collectively known as presumptive removal. From: https://en.wikipedia.org/wiki/Wikipedia:Presumptive_removal_of_AI-generated_content
NVIDIA CEO Jensen Huang doesn't want "well-respected people" like PhDs and CEOs to criticize AI, especially in front of governments. He labeled concerns about the AI technology a "doomer narrative": https://80.lv/articles/nvidia-ceo-doesn-t-want-well-respected-people-to-criticize-ai The screenshot includes a color photo of Jensen Huang wearing his typical leather jacket. He is an Asian man, with glasses and silver hair. His overall affect is smug.