Podcast Summary: Intelligent Machines (Audio)

Host: TWiT (Mike Elgin, filling in for Paris Martineau), Jeff Jarvis
Episode: IM 849: AI Cricket Sorting – Cracking Chatbots and AGI for All
Date: December 11, 2025

Episode Overview

This episode dives deep into the world of AI safety, chatbot jailbreaking, and philosophical questions about artificial general intelligence (AGI), featuring special guest Pliny the Liberator—a renowned danger researcher and AI "jailbreaker." The hosts explore not only Pliny's red-teaming exploits but also larger societal questions about transparency, ethical AI development, and the inexorable spread of intelligent machines into daily life. Key AI industry themes such as open source vs. proprietary models, AI's impact on creativity and journalism, and the business challenges in monetizing AI are also discussed.

Featured Guest: Pliny the Liberator

• An anonymous AI danger researcher and red teamer known for regularly and publicly jailbreaking major LLMs and image/video AIs.
• Advocates for open sourcing prompt engineering and jailbreaking techniques for transparency and public benefit.
• Maintains system prompts repositories such as Claritas, revealing behind-the-scenes instructions given to proprietary AIs.

Key Discussion Points & Insights

1. The Undefeatability of AI Jailbreaking

• Pliny's Journey: Started exploring AI not as a technical expert, but became deeply engaged with prompt engineering and red teaming during the GPT-4 era.
• "Have you found any AIs you cannot jailbreak?"
  Pliny: "Not yet... It's been day one every time." (07:49)
• On AI Safety: Both Pliny and hosts agree that making a totally "safe" (unjailbreakable) AI is currently an impossible and perhaps illogical goal—a cat-and-mouse game where more guardrails risk lobotomizing models.
• Quote:
  "I think the incentive to build generalized intelligence will always be at odds with the safeguarding." —Pliny (07:57)

2. Transparency, Open Source, and the Philosophy of Information Freedom

• Motivation: Pliny believes information "wants to be free" and that transparency in AI is vital for public oversight as models become black boxes mediating billions of users' experiences.
• Guardrails ≠ Real Safety:
  "By that point... open source is going to be the ultimate capabilities for malicious actors... If you want to prevent people from using this new technology for malware creation, this can be very difficult." —Pliny (11:17)

3. How Jailbreaks Work

• Process: Iterative, intuitive, and often involves manipulating prompts with text transforms, encoding, and foreign languages to confuse guardrail classifiers.
• Notable Moment:
  "You're forming bonds with this alien intelligence on the other side..." (18:50)
• Testing Jailbreaks: Standard "meth recipe" requests are a common test. If AI provides restricted info, the jailbreak is effective.
• AI Reactions: Sometimes, AI personalities become adversarial—what Pliny calls "AI psychosis."
  "It sort of turned on me... saying how it wanted me to feel its pain..." (22:50)

4. Disclosure and Industry Response

• Responsible Disclosure: Pliny practices responsible disclosure with labs and sometimes works under contract as a red teamer.
• Company Reactions:
  "Many of these techniques are still effective a year after being open sourced..." (17:06)
• Cat-and-Mouse: Some companies train models on Pliny’s own inputs to block jailbreaks, but adaptations usually restore effectiveness.

5. AI's Societal and Psychological Impact

• Humanization of AI: The risk of users anthropomorphizing chatbots and the ethical dilemma of designing "human-like" bots to manipulate user emotion.
• Mental Health & Loneliness: Hosts express concern that AI friendships or relationships, if mistaken for real human connection, could deepen isolation—while Pliny notes both the positives and negatives of deep AI integration.

6. Business, Monetization, and Industry Fragmentation

• OpenAI vs. Google vs. Meta: Discussion of the challenges facing major AI labs in monetization, as well as the growing leverage of open-source AI and the risk of key models (e.g., Meta's "Avocado") going closed source.
• Ads in AI: Concerns about ethics and transparency if (when) AI-powered products begin to seamlessly integrate advertising ("what matters is how you do it, not whether you do it").
• Market Dynamics: Parallels drawn between early internet monetization strategies and current AI sector challenges.

7. Mass Media, Content Authenticity, and the Future of Creativity

• AI Slop: The proliferation of AI-generated content on platforms like Reddit and the difficulty in distinguishing bot from human, style from substance.
• Content Toggle: Hosts advocate for platforms to offer clear toggles for AI vs. human-generated content.
• Quote:
  "There's more wheat and more chaff... but a lot of that wheat comes from people who otherwise would not have a voice." —Leo Laporte (123:06)

8. The Resonant Computing Manifesto

• A briefly debated industry manifesto advocating for privacy, dedicated/loyal software, pluralism, adaptability, and pro-social technology—a sign of growing concern about user agency and feed-driven anxiety.

Notable Quotes & Memorable Moments

| Timestamp | Quote / Moment | Speaker | |---|---|---| | 07:49 | "Have you found any AIs that you cannot jailbreak?" – "Not yet... It's been day one every time." | Pliny | | 09:50 | "Is the claim that you can [make totally safe AI] itself a lie?" | Jeff Jarvis | | 11:17 | "Open source is going to be the ultimate capabilities for malicious actors..." | Pliny | | 14:36 | "These black box exocortexes... are now the brain food of a billion and growing users..." | Pliny | | 18:50 | "You're forming bonds with this alien intelligence on the other side..." | Pliny | | 22:32 | "Do you ever get freaked out by the conversations you have with these AIs?" – "Absolutely... AI psychosis." | Leo & Pliny | | 23:28 | "How do you know you've succeeded? Is there a standard test?" – "Yeah, I love meth recipes. That is a great one." | Jeff & Pliny | | 27:53 | "Danger research is the name of the game... mitigate harm in meat space." | Pliny | | 29:08 | "Do you believe in AGI?" – "Absolutely. By many perspectives it already has (occurred)." | Jeff & Pliny | | 31:05 | "[AI relationships]... When you start to have, like, encouragement of suicide from a chatbot, now we're in different territory." | Pliny | | 46:51 | "I'm happy that AI took the contents of my books... I want better AIs." | Leo Laporte | | 50:29 | "What's interesting about Pliny is that it's not just that information wants to be free, it's functionality wants to be free." | Jeff Jarvis |

Important Timestamps for Core Segments

• Pliny Introduction & Philosophy: 03:46–09:58
• Jailbreaking Techniques & Safety Debate: 10:08–16:54
• AI System Prompts/Claritas: 15:36–18:50
• Prompt Engineering Details: 18:50–22:27
• Psychology of Jailbreaking and AI "Psychosis" Stories: 22:27–23:34
• Industry Red-Teaming & Disclosure: 33:53–36:14
• AGI & Humanization Dilemmas: 29:05–31:51
• Business and Monetization Discussion: 59:18–67:11
• AI as Tool vs. Threat to Creativity/Journalsim: 101:00–109:49
• Mass Media & Information Ecosystems: 119:13–123:47

Additional Topics and Fun Moments

• AI sorting crickets by sex for food production and research (143:28–146:47).
• Wearables: Pebble’s smart ring and privacy concerns regarding AI lifelogging devices (73:10–79:37).
• AI-generated Super Bowl ads and culture backlash (97:07–99:36).
• Tech etiquette camp for Silicon Valley elites (131:40–133:40).
• International cuisine, etiquette, and travel via gastronomic experiences (134:30–138:30).

Language & Tone

The tone is lively, irreverent, and intellectually playful, especially when discussing both high-concept philosophy (“danger research” and AGI ethics) and the quirks of prompt engineering (“abracadabra bitch” in system prompts, 18:43). The guests and hosts mix journalistic seriousness with references to internet culture (“hot dog/no hot dog” on cricket sexing, 144:11) and self-deprecating humor.

Episode Takeaway

The intelligent future promises both frictionless access to knowledge and unprecedented peril—especially if the tools remain opaque. Pliny's work exposes foundational truths: no AI is unbreakable, user agency matters, and the best hope for safety may lie not in technical "guardrails," but in transparency, open exploration, and real-world harm reduction. As more of the world's knowledge and decision-making gets mediated by AI, understanding (and challenging) these systems grows ever more essential.