Benjamin Riley
LightNews LightNews
banner
benjaminjriley.bsky.social
Benjamin Riley
@benjaminjriley.bsky.social
7.1K followers 990 following 4.5K posts
Founder of Cognitive Resonance, a new venture dedicated to helping people understand human cognition and generative AI. Advocate for humans.

Newsletter: https://buildcognitiveresonance.substack.com/
Posts Replies Media Videos
benjaminjriley.bsky.social
I will keep saying it: It turns out that inventing a tool that counterfeits the experience of being human can have, or rather is having, adverse consequences. Shirky concludes with the same point I’ve been making, that we cannot wait decades to get ahead of this. We are warping a generation.
Just as overreliance on calculators can weaken our arithmetic abilities and overreliance on GPS can weaken our sense of direction, overreliance on A.I. may weaken our ability to deal with the give and take of ordinary human interaction.
January 31, 2026 at 12:23 PM
benjaminjriley.bsky.social
Impressive work on the part of the Dartmouth admin comms team to manifest in human form the very same ethical and philosophically issues that plague generative AI.
The Office of Communications team made and suggested edits to Roberts’s draft “like we were all writing it together,” Roberts told The Dartmouth in an interview. While no statements in the piece were untrue, they were not entirely aligned with “deeper things going on that I was really feeling,” he explained.
January 29, 2026 at 11:55 PM
benjaminjriley.bsky.social
What’s going on with that new paper suggesting LLMs can make sense of Jabberwocky language? @eryk.bsky.social nerds out on it while making a nuanced defense of the stochastic-parrot metaphor.

(Link to follow)
What frustrates me in the ongoing conversation about AI is this: we continue to have this weird argument over the limits of stochastic parrot metaphors, mostly because of misplaced fixations on what parrots do. The fundamental mechanism it describes is appropriate and correct, but what emerges from that mechanism can still be complicated and deserving of attention. The fixation on proving LLMs are conducting something "akin to thought" could just as easily be framed as questions about why it is capable of resembling thought, and this paper raises some interesting questions in that arena – questions that remain interesting even if we dismiss the idea of thinking machines in favor of thinking about them.
January 25, 2026 at 1:46 PM
benjaminjriley.bsky.social
“Trump Administration conspired to violate Constitution, take political prisoners” is the headline you meant to write, New York Times.
NYT Headline: D.H.S. Cited Foreign Students’ Writings and Protests Before Their Arrests Documents unsealed by a federal judge on Thursday include dossiers that investigators prepared on pro-Palestinian student activists before they were targeted for deportation.
January 23, 2026 at 2:03 PM
benjaminjriley.bsky.social
Students know what’s up. This sharp seventh grader raises the obvious question about what teacher use of AI implies for student usage.
AI. This one is a bit of the odd one out. As students, we are forbidden from using AI for school-related purposes. However, I see my teachers using it to create assignments, mistake-filled example essays, lesson plans, and class discussion questions. some more than others. It makes me wonder: why am I spending so much time doing this assignment that was obviously created by ChatGPT or Claude (there’s literally a tab with a ChatGPT icon in the teacher’s browser!)? I wouldn’t say that AI is necessarily bad at creating assignments. But there’s sometimes factually incorrect things, or questions that don’t make sense, or analysis that feels far-fetched in a class discussion. If the students can’t use AI, why is there a double standard for the teachers?
January 19, 2026 at 8:37 PM
benjaminjriley.bsky.social
January 13, 2026 at 1:56 AM
benjaminjriley.bsky.social
NYT managing editor: "So, godlike AI system, what do we mean by this?"

NYT reporter: "“To one who has faith, no explanation is necessary. To one without faith, no explanation is possible.”

[beat]

NYT managing editor: "Make sure there's a period after the A and the I. Ship it."
Mark Zuckerberg, Meta’s chief executive, asked top executives last year to make cuts to their 2026 budgets while he pours money into A.I. research. As Meta faces competition from companies like OpenAI and Google, Mr. Zuckerberg has increased the budget for TBD Lab, the skunk works unit at Meta that aims to build superintelligence, a godlike A.I. system.
January 12, 2026 at 11:04 PM
benjaminjriley.bsky.social
I’m a bit biased since they ran my essay last month, but I’m telling you, subscribing to @theverge.com is worth it for @lopatto.bsky.social’s stories alone.
I never want to hear any moral grandstanding from these boys ever again. The next time Tim Cook says “privacy is a human right,” the only possible response is to laugh in his face. I mean, Apple and Google are fine distributing an app that has created an undressed image Grok made of Renee Nicole Good, the mother who was shot by ICE in Minneapolis. How do you plan to defend getting rid of the ICEBlock app while allowing X to generate degrading images of a woman ICE killed? Can Apple and Google even identify their values beyond their commitment to “shareholder value”? What’s your fucking endgame here, guys?
January 12, 2026 at 9:54 PM
benjaminjriley.bsky.social
I’m reading @philipcball.bsky.social’s The Book of Minds and I’m afraid I must report that cats are out here catching strays.
January 11, 2026 at 3:17 PM
benjaminjriley.bsky.social
A comment to this post suggests that local Philly activist Sam Rise led this powerful and haunting example of collective voice. Here is what they said in a recent interview about the power of art and music to make revolution irresistible:
When we make music together in a fraught time, we say we’re not disposable to one another. We’ll move through our anger, we’ll move through our grief, we’ll move with our joy. We’ll honor it all. The more my heart breaks inside the world and in relationship to the great unraveling that we’re seeing right now, there’s also this demand to bring our art, our radical imaginations, to it. That’s the one tool that I have, but it’s a powerful one. It’s one that works. It’s hard for me to imagine a world without it.
January 11, 2026 at 1:23 PM
benjaminjriley.bsky.social
January 10, 2026 at 5:43 PM
benjaminjriley.bsky.social
Anti-ICE protect, Austin City Hall - January 10, 2026.
January 10, 2026 at 5:36 PM
benjaminjriley.bsky.social
From White Rose pamphlet #3, authored by German students living under Nazi fascism.
Every individual human being has a claim to a useful and just State, one which secures the freedom of the individual as well as the good of the whole. For, according to God’s will, man is intended to pursue his natural goal, his earthly happiness, in self-reliance and self-chosen activity, freely and independently within the community of life and work of the nation. But our present “State” is the dictatorship of evil. “Oh, we’ve known that for a long time,” I hear you object, “and we don’t need to have it brought to our attention yet again.” But, I ask you, if you know that, why do you not bestir yourselves, why do you allow these men in power to rob you step by step, openly and in secret, of one domain of your rights after another, until one day nothing, nothing at all will be left but a mechanized State system presided over by criminals and drunks?
January 9, 2026 at 12:06 AM
benjaminjriley.bsky.social
Hardly the most important matter right now, but here again we can behold the mighty powers of artificial general intelligence at work, creating an apartment floorplan with an entry way that leads straight into a wall, and a desirable kitchen box.

(Link below to WaPo story)
January 8, 2026 at 3:02 PM
benjaminjriley.bsky.social
Know our history. Know the power of bearing witness to injustice to galvanize righteous protest and transformative change. It’s never been more important than right now.

meridianstar.com/2020/08/31/s...
Emmett Till’s mother, Mamie Till-Mobley, wanted the world to see “what they did to my baby.” His body looked monstrous, as if the 14-year-old had absorbed every blow of hate delivered by his killers — a photograph that ran in Jet magazine and many other African-American publications, but never appeared in the nation’s mainstream publications. Read more at: https://meridianstar.com/2020/08/31/see-the-photo-emmett-tills-mother-wanted-you-to-see-the-one-that-inspired-a-generation-to-join-the-civil-rights-movement/ In his book, Eyes on the Prize: America’s Civil Rights Years, 1954-1965, Juan Williams concluded that decision by Till’s mother “without question … moved black America in a way the Supreme Court ruling on school desegregation could not match.” Weeks later, she testified at the September 1955 trial of the killers, J.W. Milam and Roy Bryant, who laughed and lit up cigars after the all-white jury acquitted them. Read more at: https://meridianstar.com/2020/08/31/see-the-photo-emmett-tills-mother-wanted-you-to-see-the-one-that-inspired-a-generation-to-join-the-civil-rights-movement/
January 7, 2026 at 6:33 PM
benjaminjriley.bsky.social
Hell yeah.
January 5, 2026 at 2:32 AM
benjaminjriley.bsky.social
I am, once again, struck by the poverty of examples of how AI will meaningfully improve student learning. If anyone thinks these relatively simplistic tasks are the key to education revolution, I invite them to join me on a tour of literally any American school district to reassess their position.
Ms. Arnardottir, a business and entrepreneurship teacher, recently used Claude to make a career exploration game to help her students figure out whether they were more suited to jobs in sales, marketing or management. Ms. Gylfadottir, who teaches English, said she had uploaded some vocabulary lists and then used the chatbot to help create exercises for her students. “I have fill-in-the-blank word games, matching word games and speed challenge games,” Ms. Gylfadottir said. “So before they take the exam, I feel like they’re better prepared.”
January 2, 2026 at 2:24 PM
benjaminjriley.bsky.social
This Tribune story, unlike so many from mainstream press outlets, does not mince words, and the authors do not pretend to hide behind the “objective” wall. This is how reporting on fascism must be carried out.
What happened here for more than two months is unlike anything in recent American history: the federal government sending agents dressed for war into neighborhoods of the country’s third-largest city to arrest mostly people who look Latino and to ask questions later. To target people largely on the basis of their skin color, on the presumption that they may be in the country without documentation, or that they may have a criminal record, or an association with a gang.
December 28, 2025 at 3:43 PM
benjaminjriley.bsky.social
December 18, 2025 at 2:51 PM
benjaminjriley.bsky.social
In their ongoing quest to shove AI down my throat, just got an email from Google informing me that Gemini being integrated into Google Home. It contains exactly one example of generative AI working its magic wonders as this woman asks for a gift suggestion for a friend.

Sure, thing.
Picture of a woman with an infant in her lap asking her Google home device gift for a gift suggestion for a friend, because of course that’s totally when a normal person would start such an inquiry. The transcript of the response reads “Sure comma thing.” It’s all so dumb.
December 16, 2025 at 10:29 PM
benjaminjriley.bsky.social
Imagine that.
John Howard, the prime minister during the Port Arthur massacre, said in a 2016 interview with ABC Australia that observing American culture led him to conclude that "the ready availability of guns inevitably led to massacres." He added: "It just seemed that at some point Australia ought to try and do something so as not to go down the American path."
December 15, 2025 at 11:21 PM
benjaminjriley.bsky.social
I'm currently beta testing an AI literacy program aimed at students and...I just don't know, I just don't know.
December 11, 2025 at 9:37 PM
benjaminjriley.bsky.social
In a delicious irony, if you ask Time.com's annoying AI widget how to make it (the widget) go away, it recommends doing things you cannot do, or else just leaving Time.com entirely. Which, as it turns out, is not a bad way to handle AI intrusion in general.
Prompt to Time.com AI: How do I turn you off? Response: I don’t have a power switch you can flip from here. To stop interacting with me you can: Close this browser tab or window. Close or hide the chat widget on the page (the X or minimize button). Navigate away from TIME.com or reload a different page. Disable site notifications in your browser settings or sign out of your account if you’re signed in. If you’d like me to stop replying in this conversation, say so (for example, “stop” or “end”) and I will not send further messages. Would you like me to stop now?
December 11, 2025 at 2:42 PM
benjaminjriley.bsky.social
They conclude with a 16-point list of actions and safeguards they are demanding that AI companies undertake. Here's a sample:
. Perform reasonable and appropriate safety tests on your GenAI models to ensure the models do not produce potentially harmful sycophantic and delusional outputs, prior to offering the models to the public. 3. Use well-documented recall procedures with provable track records of success48 to recall generative AI products, including chatbots, if you cannot stem dangerous sycophantic or delusional outputs. 4. Provide clear and conspicuous warnings—which are permanently viewable on the same screen that a person provides inputs for your GenAI products—concerning unintended and potentially harmful outputs that may be generated by your GenAI products. 5. Develop and maintain policies and procedures that have the purpose of mitigating against dark patterns in your GenAI products’ outputs.
December 11, 2025 at 1:15 PM
benjaminjriley.bsky.social
The demand is that AI companies take measures to end this behavior. Refreshingly, however, they aren't about to be snowed by promises the AI companies make but cannot keep.
By way of example, GenAI developers have moved fast to incorporate reinforcement learning from human feedback (“RLHF”) to train their GenAI products. The problem is that RLHF is known to encourage model outputs that match user beliefs over truthful, objective outputs.39 Giving RLHF too much influence in a GenAI model’s output (e.g., via rewarding short-term feedback from thumbs-up and thumbs-down user data) can cause GenAI outputs to become more sycophantic in ways unintended by the developer, including validating users’ doubts, fueling anger, urging impulsive actions, or reinforcing negative emotions.40 GenAI developers admit this kind of unintended sycophantic behavior from GenAI products can raise safety concerns—including around issues like mental health, emotional over-reliance, and risky behavior
December 11, 2025 at 1:08 PM