Moshe Vardi

What I am about to describe ain’t AGI; it’s a sign of a trillion dollar trainwreck. If I had told you in 2022 that the 2026 version of GPT (which by the way would only be GPT 5.5 and not GPT-6 or 7 like many people fantasized about) would still have strange quirks like inserting the word “goblins” in random places, y’all would have called me either “crazy” or “a hater” or both. “Scaling”, you would have shouted. “Deep learning is conquering walls!”, you would have said. And yet here we are. OpenAI can’t even align their systems well enough to get them to stop talking about goblins without putting a bunch of utterly hack-y goblin-specific crud in their system prompts like (and I am not making this up) “never talk about goblins, gremlins, raccoons, trolls, ogres, pigeons, or other animals or creatures unless it is absolutely and unambiguously relevant to the user’s query." Meanwhile, this nonsense varies by “persona”. An actual quasi-scientific report on their website reports, without humor, “Across all datasets in the audit, the Nerdy personality reward showed a clear tendency to score outputs to the same problem with “goblin” or “gremlin” higher than outputs without, with positive uplift in 76.2% of datasets.” Instead of actual computer science, we are left with alchemy. Might as well be chanting magic incantations. Good luck solving AI safety with this tech. 🙄

As a member of the plaintiff class in the landmark $1.5 billion Bartz v. Anthropic copyright settlement — and as an investigative journalist who’s reported on many class actions over the years — I’m watching this historic deal with a close, personal eye. Anthropic settled claims that it downloaded hundreds of thousands of pirated books from shadow libraries like LibGen to train its Claude AI models. Several of my books are part of this class. Like thousands of other authors and rightsholders, I filed a claim. The proposed payout is roughly $3,000 per work. Not a windfall, but a meaningful recognition that our copyrighted material was used without permission to help build a multi-billion-dollar company. The claims rate is very high — over 90% — showing how widespread the sense of harm is among creators. Importantly, this settlement does not grant Anthropic a future license; we keep our full rights. That’s a solid outcome compared to the risks of trial on these still-developing AI copyright issues. As a non-practicing attorney, I understand why contingency counsel deserve fair compensation for taking on massive discovery, class certification fights, and real litigation risk in uncharted legal territory. Without them stepping up, many of us would still be fighting for some semblance of justice. But here’s the hitch the federal judge rightly highlighted by delaying final approval this past Friday: class counsel is seeking hundreds of millions in fees (initially in the $320M range before scaling back under pressure), which works out to effective rates of $10,000–$12,000+ per hour. I’ve covered too many class actions where the lawyers walked away rich while the “victims” they represented got little. “Fair” and “reasonable” must apply to everyone in the settlement — especially the creators whose words helped fuel these transformative AI models. This case sets important precedent for the entire creative industry as AI companies continue to ingest copyrighted material. We want the settlement approved, but with adjustments that put more money directly into authors’ and rightsholders’ pockets and send a stronger signal against unauthorized scraping. Read more arstechnica.com/tech-policy/20…

From @WSJFreeEx via @WSJOpinion: Nicholas Kristof’s column on Israeli misdeeds relies a patchwork of omissions, dubious sources and ever‑more lurid allegations, writes @rachelodonoghue on.wsj.com/4dbWbDQ

After surviving 133 years, Princeton’s Honor Code “finally met its match in generative AI,” @rosehorowitch writes: theatln.tc/KHrjLqEu 📸: Sergio Amiti / Getty

American science is at extraordinary risk. NIH has awarded less than half as many grants as it has compared to the past five fiscal years averaged together. 'I thought we were at rock bottom', the official said. 'We are below rock bottom now.'"

Trump Administration Weighs $1.7 Billion Fund for Allies Investigated Under Biden nytimes.com/2026/05/15/us/…

“That’s why the students are booing, I think. They’re experiencing AI in realtime as a forecloser of futures; as the cruel new face of hyper-scaling capitalism, as the prime agent moving a world that’s become a deck stacked against them.” — @bcmerchant bloodinthemachine.com/p/ai-as-the-ne…

Students Are Learning Less and Getting Higher Grades Because of AI, Study Finds | Ece Yildirim, Gizmodo The booming use of generative AI by students is leading to rising grade inflation at universities, according to a working paper published this week by the University of California, Berkeley. There are three ways generative AI can be used by students: augmentation, where the tools perform a supporting role assisting in things like research while the student completes the bulk of the work themselves; reinstatement of new AI-based tasks; or through displacement, where it completely automates the work that a student would otherwise perform themselves, such as writing an essay. All three use cases can improve grades, while only augmentation and reinstatement can further correlate with actual learning and skills building. Some academic tasks, like unsupervised take-home assignments, essays, and other homework, are perfect opportunities for AI displacement, as opposed to proctored exams, oral presentations, or in-class discussions. As part of the study, UC Berkeley senior researcher Igor Chirikov analyzed over 500,000 student-course enrollments across 84 departments at a large Texas university from 2018 to 2025. He found that grade increases were mostly concentrated in courses “with higher shares of writing and coding tasks,” where take-home assignments carried the most weight, concluding that students are using AI to cheat on some schoolwork and get better grades. Overall, the researchers found that “AI-exposed courses” saw a 30 percent increase in “A” grades since ChatGPT hit the market. That’s not particularly shocking; it’s a generative AI use case as old as the dawn of ChatGPT. Plus, a student’s GPA could be make-or-break for their future, determining acceptance into postgraduate academic programs and lucrative early-career job opportunities. So, in a world where most industries are leaning into AI often at the expense of the young graduate job market, it makes sense that the average student would seek out an easy way to guarantee their future. What is interesting is that, four years into the widespread presence of generative AI in our daily lives, the study shows that American universities have yet to catch up with its consequences. With more AI-enabled grade inflation, employers will have a tougher time weeding out strong young graduate candidates, the study says. But even more importantly, this increased reliance on AI in academia is bound to create an incompetent workforce that is dependent on AI. “If AI displaces skill-building tasks during learning, students may graduate with weaker capabilities in precisely the domains where AI is strongest, reinforcing a feedback loop between AI in education and AI in production that could accelerate automation,” Chirikov writes. So an academic system that caters to AI-enabled grade inflation would create a workforce that does not know how to perform the core duties of their jobs, which in turn would create increased reliance on AI in the workforce and even more wholesale automation of jobs, on the road to a much-feared AI jobs armageddon that some experts claim is already underway in some industries. Some universities are planning to take action against this grade inflation, though whether the planned measures will be truly successful is up for debate. At Princeton, where roughly 30% of seniors admitted to cheating mostly via generative AI in a recent survey, faculty voted this week to overturn a 133-year-old honor code that allowed students to take in-person exams without a faculty member proctoring. Meanwhile, at Harvard, faculty are voting on a proposal to cap A grades to no more than 20% of the class. gizmodo.com/students-are-l…

Yes, the Nakba is still "ongoing." The humiliation of 7 Arab armies failing to annihilate a newly formed state -- created by lowly Jews -- is till ongoing, on a daily basis. Once we understand the failing that Palestinians call Nakba it behooves us to rejoice it's ONGOING. Millions awe their lives to this ONGOING failing.

The roles AI replaces will not wait for graduates to retrain. A Goldman Sachs report estimated that 300 million full-time jobs globally are exposed to AI automation, with two-thirds of U.S. occupations facing some degree of impact. These college degrees are losing their economic power faster than universities can update their syllabuses: forbes.com/sites/jodiecoo…

Science publishing giant Elsevier has joined the dozens of firms and individuals suing artificial intelligence companies over their alleged use of copyrighted works in training AI models go.nature.com/4uUldNO

As New York City public schools lean on AI apps to teach students reading and math, a growing cohort of parents views it as parasitic — siphoning their children’s intellectual curiosity, capacity for boredom, and data in order to make money. nymag.visitlink.me/6xVIzx

Researchers provide "conservative estimate" of nearly 150,000 hallucinated citations in 2025 alone. arxiv.org/abs/2605.07723

“A.I. is a force multiplier of mediocrity.” newyorker.com/culture/infini…