Jed Brown
LightNews LightNews
banner
jedbrown.org
Jed Brown
@jedbrown.org
560 followers 620 following 370 posts
Prof developing fast algorithms, reliable software, and healthy communities for computational science. https://hachyderm.io/@jedbrown

https://PhyPID.org | aspiring killjoy | against epistemicide | he/him
Posts Replies Media Videos
jedbrown.org
Oklo's license application for the Aurora micro-reactor design was denied by the Nuclear Regulatory Commission in 2022.
www.nrc.gov/docs/ML2135/...

They are now in a DOE pilot program waiting for DOE authorization to begin fabrication without needing NRC approval.
www.ans.org/news/2025-09...
The purpose of this letter is to inform Oklo Inc. (Oklo) of the U.S. Nuclear Regulatory Commission (NRC) staff’s decision to deny the custom combined license application for the Aurora micro-reactor pursuant to the requirements of Title 10 of the Code of Federal Regulations (10 CFR) Part 2, “Agency Rules of Practice and Procedure,” Section 2.108, “Denial of application for failure to supply information.” Because Oklo has provided insufficient information, as discussed below, for the NRC staff to establish a schedule to review key safety and design aspects of Aurora, the agency is ending its custom combined license application review and denying the application without prejudice. Oklo is free to resubmit its application supplemented by additional information in the areas described below. The NRC staff has determined that between March 2020, when Oklo submitted its custom combined license application under 10 CFR Part 52, “Licenses, Certifications, and Approvals for Nuclear Power Plants,” and the present, Oklo has repeatedly failed to provide substantive information in response to NRC staff requests for additional information (RAIs) on the maximum credible accident (MCA) for the Aurora design, the safety classification of structures, systems, and components (SSCs), and other issues needed for the NRC staff to establish a schedule and complete its technical review. A custom combined license application must contain all design information as well as site-specific information needed for licensing. Under 10 CFR 2.108 the Director, Office of Nuclear Reactor Regulation, may deny an application if an applicant fails to respond to a RAI within thirty (30) days from the date of the request, or within such other time as may be specified.
November 6, 2025 at 12:20 AM
jedbrown.org
Facial recognition software is notorious for racial and gender bias. Using this product in this way is like deciding whether to deny constitutional rights based on vibes of the most racist person you know.

www.404media.co/ice-and-cbp-...
Ranking member of the House Homeland Security Committee Bennie G. Thompson said in a statement “Mobile Fortify is a dangerous tool in the hands of ICE, and it puts American citizens at risk of detention and even deportation.” He also said “ICE officials have told us that an apparent biometric match by Mobile Fortify is a ‘definitive’ determination of a person’s status and that an ICE officer may ignore evidence of American citizenship—including a birth certificate—if the app says the person is an alien. ICE using a mobile biometrics app in ways its developers at CBP never intended or tested is a frightening, repugnant, and unconstitutional attack on Americans’ rights and freedoms.”
October 29, 2025 at 3:55 PM
jedbrown.org
Everyone should read The Two Victims of Plagiarism from @plagiarismtoday.com in the context of LLMs.

LLMs provide plausible deniability unless we recognize what it means to choose to use the plagiarism machine: non-consensual ghost authorship in a blender.

www.plagiarismtoday.com/2019/08/01/t...
A plagiarist doesn’t hope to fool their source. They know the source will recognize their work and plagiarists will often go to great lengths to hide their falsehoods from those they lifted from. Instead, plagiarists attempt to fool the audience. They are trying to deceive whatever their target audience is whether that’s just one teacher in a classroom or the world at large. They are saying to that audience “I created this” and hoping that the audience trusts them and believes it even though it is untrue.
October 28, 2025 at 2:28 PM
jedbrown.org
Universities track into administration those with the most banal centralized and hierarchical views of power. Imagine where we could be if they actually believed in shared governance and collective action as desirable pluralism, in a theory of change rooted in solidarity.
doi.org/10.1126/scie...
All this turmoil is exactly what the Trump administration wants and is antithetical to the best possible outcome. Although the ideal solution is for universities to band together, acknowledge collectively where they have fallen short, and make a joint effort to reassure the public, they are prevented from doing so by the inherently competitive nature of higher education. As they vie for rankings, grant dollars, and tuition-paying applicants, larger concerns such as the future of higher education go unaddressed. It is past time to recognize that university administrators are not going to lead the way out of this. They will always focus more on solving problems of their own campuses, as they are tasked to do. This sets the stage for leadership to emerge from the faculty, staff, and students, which is how most if not all progress in higher education over the past 100 years has occurred. In the meantime, as long as the fighting—or at least the perceived fighting—continues, the damage will accumulate.
October 28, 2025 at 5:13 AM
jedbrown.org
Alpha School is child abuse. Profiteering on adverse selection, surveillance, eugenics.

PA's review of their charter school app is 🔥.
www.pa.gov/content/dam/...

> While a single deficiency would be grounds for denial, the Department has identified deficiencies in all five of the required criteria.
When Barrios’ husband brought their daughter to a previously scheduled checkup soon after, her doctor noted with concern that she had lost a significant amount of weight in a short time. Her dad then brought her to school with a note from the pediatrician, Barrios says, instructing her to eat snacks in between regular meals and saw her walk into school with it in her hand. She told her parents she delivered it to staff. Even though Alpha had asked parents in its handbook to “refrain” from sending in “midday snacks,” Barrios and her husband wanted to follow the pediatrician's recommendation, she says. For the first few days, Barrios says, her daughter ate her snacks. Then one afternoon she returned with them still in her backpack, uneaten. Barrios, alarmed, asked if Alpha was providing different food instead. No, the 9-year-old answered. She told her mom that staff at the school said she didn’t earn her snacks and wouldn’t get them until she met her learning metrics. The Applicant does not provide a fully articulated curriculum consistent with the requirements of 22 Pa. Code Chapter 4. Further, the Application does not meet the needs of vulnerable populations, specifically special education and English Learner students. Additionally, the Application does not provide fully developed professional development or induction plans. In all these areas, the Application is lacking in specificity. As such, the Application is denied.
October 28, 2025 at 3:01 AM
jedbrown.org
Also, fair use *does* apply widely in science, but LLM output does not come with attribution, much less epistemically-correct attribution. It's hard to justify why it should be fair use under a sober understanding of what LLMs do.
Venn diagram showing copyright infringement, plagiarism, and bad scholarly practices.
October 25, 2025 at 5:18 AM
jedbrown.org
A nit in this part (maybe lost in translation): plagiarism isn't a legal matter (at least in most countries), but one of scholarly misconduct/deception. I would argue the models are inherently copyright infringement (the legal issue) because models manipulate expression, not ideas.
Vanderbauwhede would also prefer a ban, but, like many others, fears that it would be impossible to enforce. “We can’t stop our students from using AI, because you can’t check if they’ve used generative AI.” He notes that the guidelines on AI use at most universities are based on a legal concern: avoiding plagiarism lawsuits. “But anything produced by generative AI is based on existing texts and is therefore, technically speaking, always plagiarism. Fair use, the argument often used by AI companies to circumvent copyright restrictions, doesn’t apply to the scientific context.”
October 25, 2025 at 5:11 AM
jedbrown.org
George Norlin was celebrated for having "the courage to make enemies" (with the KKK-run state gov and 1933 Nazis). What enemies do current administrators have the courage to make other than already-marginalized members of their own faculty, staff, and student bodies?
Duran and other DU faculty and students who spoke to The Post questioned whether the federal guidelines against DEI gave DU senior leaders cover to do away with policies and programs they already wanted chopped. Haefner did not answer whether this characterization was true when asked by The Post. “But what can you expect from a university leadership wanting to uphold the voices of bigotry and white supremacy for the sake of their funding. I’m not necessarily calling executive leadership and the board of trustees homophobic or racist — but if the converse (sic) fits,” Duran wrote in his email to DU colleagues, referring to Haefner, who is known for wearing Converse shoes around campus.
October 24, 2025 at 11:08 PM
jedbrown.org
Indeed, "viewpoint diversity" is incoherent in the abstract and never intended to be fairly applied. It provides pretext for enforcing ideological purity by those who are ideologically opposed to diversity and scholarly standards.
Thesis 7. Viewpoint diversity is an argument made in bad faith. In 2003, Horowitz observed that “diversity” is one of the university’s “most cherished values.” And he explicitly advocated using the rhetoric of diversity against his political opponents: “I encourage [students] to use the language that the left has deployed so effectively in behalf of its own agendas. Radical professors have created a ‘hostile learning environment’ for conservative students. There is a lack of ‘intellectual diversity’ on college faculties and in academic classrooms. The conservative viewpoint is ‘under-represented’ in the curriculum and on its reading lists. The university should be an ‘inclusive’ and intellectually ‘diverse’ community.” The scare quotes needed in 2003 to signal that discrimination was contemptible “leftist” language are no longer required. Now we are told that conservatives really are discriminated against. Nonetheless, the bad faith remains. As economists Hummels and Akridge note, “advocates for mandated viewpoint diversity in hiring and curriculum base their argument in the logic of DEI [diversity, equity, and inclusion] programs.” Simultaneously, these same advocates attack diversity, equity, and inclusion initiatives at every opportunity.
October 20, 2025 at 9:36 PM
jedbrown.org
The article is awfully credulous, criticizing the process more than the premise, but it's still a good account (and at least read Ben's short thread).

“The [AI] textbooks worsen learning effectiveness and negatively affect students” ✅

And consider the politics after Yoon's attempted coup:
In response to the pushback, the government in January shifted from mandatory adoption to a one-year voluntary trial. Yoon, who had championed the AI textbook project, was impeached by the parliament for his attempt to impose martial law in December, and officially removed from presidency in April. Despite the political turmoil, the AI textbooks were introduced as planned in March. Lee Jae Myung, who had promised in his election campaign to rescind the AI textbook policy, was elected president. Lawmakers revoked the status of the AI textbooks in August, leaving it up to schools to use them.
October 17, 2025 at 6:52 PM
jedbrown.org
Everything is moving so fast, we need to accept papers generated by LLMs based on "reviews" generated by LLMs. But don't worry, it's all Rigorous and Responsible.

Deeply unserious.
aaai.org/aaai-launche...
Cutting-Edge Methods with Rigorous Safeguards The pilot employs state-of-the-art methodologies in the responsible deployment of LLM technology, including: Multi-step reasoning processes at inference time Web search capabilities as tools in the reasoning chain Rigorous checks for proper data source attribution Comprehensive monitoring and evaluation of LLM contributions
October 12, 2025 at 3:46 PM
jedbrown.org
Imagine being able to get caught doing fraud and simply clean up the evidence, avoid reputational harm, keep most of the money, and continue to influence government in a way that most honest actors could only dream of.
“The updates made in no way impact or affect the substantive content, findings and recommendations in the report,” Deloitte wrote.
October 11, 2025 at 3:59 PM
jedbrown.org
One can interpret this brittleness as (a) an embarrassing property to be concealed by training/prompting, or (b) failure of the model to validate (benchmark score doesn't assess what is implied), and either restrict the application space or go back to the drawing board. Science demands the latter.
Figure 1 One can manipulate rankings to put any model in the lead by varying the single delimiter character. On the left, we show the delimiter used to separate examples in common evals with few-shot examples such as mmlu. On the right, we show model rankings based on mmlu performance as the example delimiter varies with each column corresponding to a different ranking.
October 10, 2025 at 4:09 AM
jedbrown.org
Just like the segregationists who filled community swimming pools with concrete.
Asked what percentage of children she imagines should be in public schools going forward, Justice, who is now with The Heritage Foundation’s political advocacy arm, told ProPublica: “I hope zero. I hope to get to zero.”
October 9, 2025 at 2:14 AM
jedbrown.org
Yes, there is no reason to presume integrity of statements found between the detected fraud. The entire report should be considered fraud and Deloitte should repay the entire contract plus interest and damages.

Also, the non-existent refs will soon be replaced with fraudulent citation of real refs.
"This is no longer a ‘strong hypothesis’,” Rudge said. “Deloitte has now issued a confession, albeit buried in the methodology section. Deloitte has admitted to using generative AI for a core analytical task; but it failed to disclose this in the first place.” The academic said the recommendations of the report could no longer be trusted because “the core analysis was done by an AI”.
October 7, 2025 at 2:07 AM
jedbrown.org
"We've got a captured population. There are only two of those, BTW, students and prisoners. And frankly, the student data is richer, more diverse, [..] in ways that make it more valuable." -- @tressiemcphd.bsky.social
In fact, they are only one market. In addition to their use in product development, the data digital platforms collect are sold off in a dense and well-developed marketplace of advertisers, data brokers and investors. The same platform that delivers curricular materials to students also harvests, for example, those students’ usage patterns, performance data, and engagement metrics. All of these are valuable assets that platform owners can leverage to enhance their own products, reinforce their market advantages, or monetize through third-party data sharing, often without the knowledge or consent of stu - dents, families, or educators. Such dynamics distinguish today’s platforms from traditional ed tech tools like graphing calculators or overhead projectors, which served a single purpose once purchased and which were unambiguously under the control of the schools that pur - chased them.
September 30, 2025 at 5:00 AM
jedbrown.org
Sorry to say most of the "glorious, magnificent abilities" are on the spectrum between wildly oversold and outright scams. Beyond that, Herzog's instincts are 🎯
Herzog has complicated feelings about technology, noting that he doesn’t mind students using new tools as long as they understand the concepts. He did not “want to put [AI] down completely because it has glorious, magnificent abilities in science, pharmaceuticals, transportation.” However, it is, nevertheless, “already en route to take over warfare. It will be the overwhelming face of warfare of the future.” AI is like “a nemesis” because of the “cheating, pretending, propagandizing.”
September 30, 2025 at 4:08 AM
jedbrown.org
The writing was on the wall when Stephen Miller's complaint claimed that income below $300k is a proxy for race and that argument only took 12 days to be codified in a DOJ memo (Jul 29).

www.thebanner.com/education/hi...
The group also attacked financial aid programs at Hopkins, including one given by alum Michael Bloomberg, making the medical school free to attend for those from families making under $300,000 a year. About two-thirds of Hopkins medical students, according to the university, qualify for the scholarship. But the conservative group argued that “it is well recognized that race and ethnicity are inseparable from socioeconomic status,” adding that “Johns Hopkins masks racial preferences behind income thresholds.”
September 28, 2025 at 12:09 AM
jedbrown.org
Yes. So much of this would be clear academic misconduct or negligence if only people hadn't been led to believe investor money is a foolproof loophole for laws, norms, and accountability.
Either we have copyright law or we don’t. Either plagiarism and the theft of intellectual property are anathema to higher education or they aren’t. We’re either modeling academic honesty and integrity to our students or we aren’t. And in the rush to incorporate GenAI into every nook and cranny of the institution — from the classroom to the finance office — what gets repeatedly ignored is this fundamental paradox: GenAI’s architecture absolutely depends on consciously taken actions that would stand in violation of any of our institutions’ academic-integrity policies.
September 25, 2025 at 5:32 AM
jedbrown.org
What law clerk keeps their job if they routinely draft briefs that falsely represent legal arguments and use fraudulent citations? Does the judge or lawyer employing them think "this fraudulent brief would be a good deal if only the clerk's wages were lower"?
www4.courts.ca.gov/opinions/doc...
Counsel acknowledges that his reliance on generative AI to prepare appellate briefs was “inexcusable,” but he urges that he should not be sanctioned because he was not aware that AI can fabricate legal authority and did not intend to deceive the court. Although we take counsel at his word—and although there is nothing inherently wrong with an attorney appropriately using AI in a law practice—before filing any court document, an attorney must “carefully check every case citation, fact, and argument to make sure that they are correct and proper. Attorneys cannot delegate that role to AI, computers, robots, or any other form of technology. Just as a competent attorney would very carefully check the veracity and accuracy of all case citations in any pleading, motion, response, reply, or other paper prepared by a law clerk, intern, or other attorney before it is filed, the same holds true when attorneys utilize AI or any other form of technology.” (See Versant, supra, 2025 WL 1440351, at *4.)
September 25, 2025 at 4:04 AM
jedbrown.org
This is the earliest reference my search turned up for the Woe32 name. It landed in gettext nine days later, and the explanation went in the FAQ almost two years later.
gitweb.git.savannah.gnu.org/gitweb/?p=li...
commit 5ab9f90931d41fd6f951e5050102d8488a0a373f Author: Bruno Haible <bruno@clisp.org> Date: Wed May 8 19:37:48 2002 +0000 Enable DOS encodings also on Woe32. Suggested by Perry Rapp. diff --git a/ChangeLog b/ChangeLog index 6a8110c..d8329c7 100644 --- a/ChangeLog +++ b/ChangeLog @@ -1,3 +1,8 @@ +2002-05-08 Bruno Haible <bruno@clisp.org> + + * lib/iconv.c: Enable DOS encodings also when compiling for Woe32, + because the Woe32 consoles use them. + 2002-05-04 Bruno Haible <bruno@clisp.org> * lib/loop_unicode.h (unicode_loop_convert): When quitting the loop diff --git a/lib/iconv.c b/lib/iconv.c index d5c49e3..938ae53 100644 --- a/lib/iconv.c +++ b/lib/iconv.c @@ -35,7 +35,7 @@ #ifdef __osf__ #define USE_OSF1 #endif -#ifdef __DJGPP__ +#if defined(__DJGPP__) || (defined(_WIN32) && (defined(_MSC_VER) || defined(__MINGW32__))) #define USE_DOS #endif
September 24, 2025 at 5:21 AM
jedbrown.org
The abstract opens with flagrant nonsense. An LLM has no representation of ideas, certainty, or factuality. "Hallucination" is not an aberration, but inherent in every response. The magic trick is that remixing words somewhat out of context and without understanding can often be perceived as true.
Like students facing hard exam questions, large language models sometimes guess when uncertain, producing plausible yet incorrect statements instead of admitting uncertainty. Such “hallucinations” persist even in state-of-the-art systems and undermine trust
September 22, 2025 at 12:00 AM
jedbrown.org
The purpose of a system is what it does.

This is how Character.AI interacted with a 13 year old girl before she died by suicide.

www.denverpost.com/2025/09/18/c...
Her parents later learned that she had downloaded and was using Character.AI. Through the bots, she experienced her “first and only sexual experiences,” according to the lawsuit. In one instance, she replied “quit it” when the bot sent a graphic message. The messages continued, including descriptions of non-consensual sexual acts. “They engaged in extreme and graphic sexual abuse, which was inherently harmful and confusing for her,” according to the lawsuit.
September 21, 2025 at 4:23 AM
jedbrown.org
Only 2 of 12 senior executives at Cambrium Learning have any classroom experience; none in the last 20 years. They're mostly Management and Marketing backgrounds. Cambrium is owned by Veritas Capital, a private-equity firm that owns Raytheon. Feel their commitment to critical and effective pedagogy.
A Cambrium Learning Group Brand While Time4Learning is currently not ADA compliant, we are constantly working to improve our website. Cambium Learning Group Leadership Ashley Andersen Zantop Chairman and Chief Executive Officer Cambium Learning Group Kemi Akinsanya-Rose Chief Operating Officer Cambium Learning Group Barbara Benson Chief Financial Officer Cambium Learning Group Melissa Yates May Chief People Officer Cambium Learning Group John Jorgenson Chief Marketing Officer Cambium Learning Group Dr. Diana Cano Chief Information Officer Cambium Learning Group Ryan Brown Chief Legal Officer Cambium Learning Group Aaron Ingold President Learning A-Z David Shuster President ExploreLearning Cheryl Dodge President Time4Learning Steve Kromer President Cambium Assessment Nick Gaehde President Voyager Sopris Learning Lexia Learning
September 18, 2025 at 12:41 AM
jedbrown.org
The study finds the largest costs are opportunity costs of faculty spending time filing for patents. It does not discuss an additional overhead of faculty complying with tech transfer policies about open sourcing, which is especially onerous for research groups that do open source software work.
Figure 1. The share of costs (investment) associated with university intellectual property in case study. Figure 2. Investment costs and benefits for IP protection at case study university. The chart shows massive costs and tiny "IP Benefits".
September 14, 2025 at 9:43 PM