WT: Going Weird Mode

Episode Notes

The episode centers on Andrew Mayne explaining his role at OpenAI and the sudden breakout of ChatGPT. He says he works on the communications team helping explain the company’s technology, and the panel discusses how ChatGPT started as a research preview, gained huge adoption quickly, and changed AI from a theoretical or magical idea into a practical tool people can try for themselves (L25, L69, L73, L81, L97, L125).

A substantial portion of the conversation is technical and reflective: they cover prompt writing, hallucinations, safety warnings, fine-tuning, context windows, and how larger models may enable more personalized or useful applications. Later, the show moves to broader questions about training data, AI agency and motivation, human meaning in art, and how technology shifts create new creative and business opportunities; the episode then closes with picks for Avatar 2, The Last of Us, Pocket Card Jockey: Ride On!, and Wednesday (L61, L125, L169, L237, L255, L337, L357, L377, L389, L429, L445).

Key topics

• OpenAI communications and Andrew Mayne’s role: Andrew says he works for OpenAI on the comms team, helping explain the company’s technologies to journalists and the public, after previously working on engineering/applied work (L25, L33).
• ChatGPT as a research preview and rapid adoption: Andrew stresses that ChatGPT was launched as a research preview, not a finished product, and the group discusses how removing waitlists, costs, and other friction helped drive explosive adoption (L69, L73, L81).
• Prompt writing and talking to machines: Andrew describes being an internal prompt writer and says that with ChatGPT, 'everybody is a prompt writer now,' while the conversation treats prompting as a real skill (L61).
• Hallucinations and safety framing: Andrew repeatedly warns that models can make things up and says OpenAI tries to be upfront about limitations and not overpromise, especially for critical use cases (L61, L125).
• Accessibility and practical use cases: Examples include helping someone with dyslexia write a story or letter, writing holiday cards in English, and helping a student communicate better across languages (L97).
• Context windows and fine-tuning: Andrew explains GPT-3 token limits, how fine-tuning works, and how larger context windows in GPT-3.5 make custom or personalized applications more feasible (L169).
• AI interfaces and structured workflows: The discussion highlights that ChatGPT’s structured interface and better formatting make it easier to use than a raw playground, especially for tasks like writing work emails or copying code blocks (L157, L161, L189).
• Training data, authorship, and scale: Brian asks about fairness in feeding human-created material into AI, and Andrew compares it to how writers are shaped by what they read while noting that scale is the key difference (L253, L257).
• Stateful AI, agency, and motivation: Andrew contrasts current stateless chat systems with future stateful systems that can keep running, self-prompt, and pursue goals, framing agency as the more important concern than consciousness (L337, L345).
• Human meaning in art: They discuss Nick Cave’s criticism of a ChatGPT-generated song and argue that AI-generated work may be technically strong but still lack the lived human meaning attached to art made by people (L357).
• Technology disruption and new creator models: Andrew uses magic books, DVDs, file sharing, and YouTube-style scale to argue that technological disruption destroys some business models while creating new opportunities (L265, L273, L297).
• Video game adaptations: The group explains why The Last of Us seems like a strong adaptation and why Warcraft is harder to translate because the emotional core of many games lies in player experience rather than lore (L397).
• Wednesday and supernatural series structure: Wednesday is praised for its world-building, casting, Tim Burton’s involvement, Christina Ricci’s role, Jenna Ortega’s choreography, and for avoiding a heavy chosen-one framework, though the final episodes’ pacing is criticized (L445, L453, L461, L469).

Picks

• Brian Brushwood: Avatar 2: The Way of Water — Strong recommendation; Brian explicitly says it is his pick and calls it a great movie and a refreshing theatrical spectacle.
• Justin Robert Young: The Last of Us — Clear pick framed as a likely strong adaptation of a video game into television.
• Bryce Castillo: Pocket Card Jockey: Ride On! — Very enthusiastic recommendation to download and play the Apple Arcade version immediately.
• Andrew Mayne: Wednesday — Positive recommendation with minor pacing criticism; Andrew says he enjoyed it thoroughly and looks forward to next season.