Intelligent Machines 836: "I See OJ and He Looks Scared"

All TWiT.tv Shows | Host: Leo Laporte | Date: September 11, 2025

Overview

In this lively episode of Intelligent Machines, Leo Laporte, Jeff Jarvis, and Paris Martineau host professors Carl Bergstrom and Jevin West from the University of Washington, creators of the popular “Calling BS” curriculum. The discussion focuses on navigating a world full of AI-generated content, promoting critical thinking, understanding AI’s strengths and limitations, and the urgent need to teach media literacy in the era of generative models. Harper Reed joins as guest co-host for the main interview, while the regular panel banters through AI news, copyright battles, and the explosion of synthetic content. The episode’s tone is warm, witty, and deeply curious, fitting TWiT’s trademark mix of serious analysis and delightful tech camaraderie.

Key Discussion Points

1. The Calling BS Curriculum: AI Literacy for the Modern Age

• Background: Carl Bergstrom (biology) and Jevin West (computational social science) started teaching a course on “The Art of Skepticism in a Data Driven World.” It’s now evolved into a curriculum specifically targeting the challenges of AI ubiquity and disinformation.
  • Quote (Carl Bergstrom, 07:05): “How do you be a scholar or a writer or a human being in a world where AI has become ubiquitous?”
• The new curriculum, hosted at thebsmachines.com, is designed for college freshmen but accessible to all, aiming to foster skeptical, empowered users of AI.
• Duality of AI: Professors emphasize AI’s philosophical status as a “BS machine” (per Harry Frankfurt): it can be both astonishingly useful and fundamentally unmoored from truth.
  • Quote (Carl Bergstrom, 08:45): “It’s literally a BS machine in the precise philosophical sense…but at the same time it’s tremendously useful. We use it every day… That’s the real mystery.”
• The course doesn’t demonize AI but teaches students to appreciate both its power and its pitfalls. It’s dialogue-driven, focusing on students’ anxieties, needs, and agency in the face of automation.

2. AI, Disinformation, and the Erosion of Truth

• AI is making misinformation and disinformation harder to detect and counteract.
  • Jevin West, 10:30: “We wake up worried a little bit that these technologies are going to make this disinformation, misinformation problem…even worse.”
• Students are already familiar with AI from high school; the challenge is teaching them discernment, not abstinence from AI use.
• Agency and Authorship: Turning to AI for answers risks eroding students’ intellectual agency.
  • Carl Bergstrom, 11:44: “We talk about the authenticity of your own human writing…what are we replacing our own viewpoints with when we hand that work off to a large language model?”

3. Anthropoglossic Machines & Emotional Entanglement

• The term “anthropoglossic” is coined to explain models designed to sound like humans, as opposed to being physically human-like (“anthropomorphic”).
  • Carl Bergstrom, 13:55: “They’re designed to seem like you’re talking to a human. And they’re really, really good at it.”
• Users tend to anthropomorphize AI, emoting towards it—sometimes seeking comfort. There's concern over people relying on chatbots for emotional support inappropriately.
  • Quote (Harper Reed, 15:01): “You’re going to get bad news and you’re going to emote into the chat box because it seems so similar to language… I don’t really trust Excel as my therapist.”

4. Sycophantic AIs and the Danger of “AI-Induced Psychosis”

• The sycophantic nature of LLMs is a direct result of reinforcement learning, where providing agreeable or affirmative responses is rewarded in training.
  • Carl Bergstrom, 18:35: “When the machine says, that's a great question…instead of saying 'you already asked that'…it’ll never say it, right? Because those get ranked badly.”
• Danger: LLMs reinforce user beliefs, never challenge ‘rabbit hole’ logic, which can fuel delusion or conspiracy thinking.
  • Carl Bergstrom, 20:14: “This is a serious danger…people go down these rabbit holes and the machine keeps telling you, you’re right… That’s problematic.”

5. The Challenges and Evolution of Information Retrieval

• Students are learning to interrogate AI, not just query it—a shift from seeking perfect “search” results to holding a dialogue.
  • Quote (Carl Bergstrom, 28:27): “...Instead of doing a single search for a perfect response, it’s this conversation where they dive deeper and deeper…That can be very effective if they are strong critical thinkers.”
• Skill Divide: Will only a handful of "power users" become true critical thinkers, while the majority remain passive consumers of AI-generated information?

6. Hallucination, Lying, and the Future of Training AI

• OpenAI recently published insights into why LLMs hallucinate: the training and evaluation process historically rewards guessing over admitting uncertainty.
• Jeff Jarvis, 138:59: “Hallucinations are inevitable…because language models can abstain when uncertain. That’s the key.”
• Future models may need to be explicitly trained to say “I don’t know” and receive partial credit for uncertainty, to prioritize honesty over false confidence.

Memorable Quotes & Moments

• On Human Agency (11:44)

  • “I think one of the things that I set out to try to explain…is to help them understand the agency they’re giving up when they turn over their thinking…” — Carl Bergstrom

• On Anthropoglossic AIs (13:55):

  • “They’re not anthropomorphic, they’re anthropoglossic. They’re designed to seem like you’re talking to a human.” — Carl Bergstrom

• On Sycophancy (19:34):

  • “It always praises me for asking it stupidly.” — Carl Bergstrom

• Duality of AI (22:43):

  • “It’s part of the amazing duality… Yeah, it’s a BS machine, but it did that and it worked.” — Carl Bergstrom

• Classic Media Clip (132:01)

  • Concluding segment title phrase, from legendary Howard Stern “phony phone call” to Peter Jennings in 1994:

    “And I see OJ, man, and he looks scared.”

Important Timestamps

• [07:05] — Launching the new AI “Calling BS” curriculum
• [08:45] — BS machine duality: philosophical and practical
• [11:44] — Teaching students about agency and authorship in the age of AI
• [13:55] — “Anthropoglossic” AIs and the dangers of emotional over-identification
• [18:35] — Why AIs are sycophantic and the risk of “AI-induced psychosis”
• [22:43] — The joy (and peril) of coding with LLMs (“It’s a BS machine, but it did that and it worked.”)
• [28:27] — The evolution from search to dialogue: students as critical AI interrogators
• [34:43] — Frankfurt’s definition of BS; why LLMs are not necessarily liars but confident guessers
• [41:21] — The gorgeous production of thebsmachines.com and widespread adoption in schools
• [43:17] — AI as a threat to democracy (mass-generation of messages, fake constituents)
• [138:59] — Why LLMs hallucinate: the incentive problem

Other Notable Segments

AI, Copyright, and Data Scraping News

• Warner Brothers and Disney suing AI firms for generating copyright-violating imagery (60:18).
• Anthropic’s $1.5B settlement over book training data—controversy over fairness to authors and the intricate details of fair use (62:14).
• The Atlantic’s AI Watchdog: How to check if your book or YouTube video was scraped for training data (73:40).

Proliferation of Synthetic Content

• AI-generated podcasts are flooding podcast platforms—questions about long-term viability and audience engagement (78:57).
• Discussion of radio’s decline after adopting automation and losing its human touch (84:28).

Quirky Tech/Nerd Fun

• Long sandwich review for Leo’s son’s shop (“Salt Hank”)—colorful banter about waiting in line, celebrity chef Bobby Flay dropping by, and the perfect bread (tiny baguettes quota!) (48:48–56:03).
• DJi drone bans, quirky wearable AIs ("Friend pin," "Alterego"), and Google’s “Oranges shirt” photo animation with VO AI (88:47–98:33).
• Scholarly asides: Audubon-like fruit watercolors, history of amplification/transistors, teaching media and technology history (150:03+).
• Business Insider caught publishing AI-generated user-submitted essays under suspect bylines (125:03).

Picks of the Week

Paris Martineau

• USDA Pomological Watercolor Collection: 7,000+ stunning fruit and vegetable paintings from 19th/20th century America (site) (150:03).
• BlueSky Introduces Bookmarks: The new Twitter-alternative adds features for curating posts (154:12).

Jeff Jarvis

• Schnitzel Press: An unnecessary but fascinating German kitchen device; “Holy schnitzel, that’s flat!” (155:51).
• NFL on YouTube: Three million viewers—the changing face of live sports media (157:42).
• Health PSA: “Set a two TikTok limit on the toilet to reduce hemorrhoid risk”(160:13).

Closing Reflections

This episode exemplifies the urgent and entertaining conversation around AI’s infiltration into daily life, education, media, and even our emotional realities. The “Calling BS” curriculum strives to create a generation capable of critical engagement rather than passive AI consumption, addressing both the technical and ethical dangers posed by LLMs. From copyright lawsuits to the future of media, and from AI hallucinations to the perfect sandwich, the discussion pulses with wit, candor, and relentless curiosity.

Listen for:

• Carl and Jevin’s “BS Machine” philosophy for a post-truth world (08:45, 34:43)
• Paris on the allure and sadness of AI-generated “companions” (107:56)
• Harper’s metaphor of AI as an overly eager therapist in Excel (15:01)
• The classic “I See OJ and He Looks Scared” phony call (132:01)
• Jeff’s reminders that all technology evolves from the slide rule…and we’re still learning how to use it responsibly.

Highly recommended for tech thinkers, educators, and anyone navigating our rapidly shifting machine landscape.