AI, Podcasts, and the Future of Creative Writing

Episode Notes

The episode opens with the hosts reflecting on how quickly AI is changing and pushing back on the idea that it will simply replace human roles. They argue that teaching, parenting, preaching, customer service, banking, and restaurant work still involve human presence, trust, empathy, and cultural meaning that AI cannot fully replace, even if it can augment or automate parts of those jobs. [L41-L49] [L53-L57] [L81-L101]

The middle of the episode focuses on OpenAI's GPT-5 and open-source local models. Andrew says GPT-5 is cheaper and more capable than earlier models and describes how OpenAI's open-source GPT OSS models can run locally on a desktop with tools like LM Studio or Ollama. The group also discusses reasoning models for creative writing, showing how better prompts and higher reasoning effort can produce stronger flash fiction and revealing the model's planning process before it writes. [L135-L149] [L155-L165] [L171-L177] [L243-L257] [L259-L273]

The latter part turns to practical AI integrations and media picks. The hosts discuss ChatGPT connectors for Gmail and Calendar, agentic workflows, always-on assistants, local speech-to-text, and the idea that AI will keep improving rather than hitting a wall. They close with entertainment recommendations and reactions: Andrew on Fantastic Four, Brian on Predator, Justin on Wednesday, and Brian and Andrew on Alien: Earth and franchise lore. [L295-L305] [L359-L377] [L387-L397] [L401-L417] [L451-L477] [L483-L493]

Key topics

• Human roles that remain hard to automate: The hosts repeatedly distinguish between tasks AI can automate and human roles that depend on trust, empathy, inspiration, or authority. Examples include teachers, parents, pastors, bankers, customer support, and restaurant servers. [L41-L49] [L53-L57] [L79-L101]
• GPT-5 and model personality differences: Andrew praises GPT-5 as cheaper and more capable than prior models, while Brian notes that GPT-4o feels playful and GPT-5 feels more sober and factual. They treat tone and personality as real user-facing differences. [L135-L149]
• Local open-source models and desktop inference: Andrew describes running GPT OSS 20B locally on a Mac Studio using tools like LM Studio and Ollama, emphasizing that a surprisingly capable model can now run on personal hardware. [L155-L165] [L259-L273]
• Reasoning models as a creative-writing tool: Andrew explains that reasoning models work best with structure, constraints, and enough context, rather than a bare 'write a story' prompt. The conversation highlights visible planning and different reasoning effort settings. [L171-L177] [L243-L257]
• Dictation and speech-to-text: Andrew mentions NVIDIA's Parakeet as a local speech-dictation model and says it can transcribe minutes of audio very quickly, treating local transcription as effectively solved. [L195-L205]
• ChatGPT connectors and agentic productivity: The hosts discuss Gmail and Calendar integrations, search behavior, and how connecting AI to personal data makes it much more useful for practical tasks like finding emails or organizing memory. [L295-L305] [L307-L349] [L359-L377]
• Browsers and interface redesign around agents: Justin argues that browsers should be rethought for an agentic era, with less friction and more direct access to information or actions, while Brian raises the importance of preserving a user's mental map. [L103-L121]
• Always-on AI assistants and memory augmentation: Andrew imagines local systems that can search files, calendars, and personal history, carrying multiple conversations and helping with autobiographical memory. [L303-L305]
• AI-generated creative outputs and immersive media: The hosts talk about how cheap, promptable generative AI could support richer interactive environments, including creative writing, NPCs, and games. [L297-L302]
• Alien: Earth worldbuilding and corporate power blocs: Brian and Andrew discuss Alien: Earth's design, pacing, and exposition about cyborgs, synths, hybrids, and rival corporations. [L471-L489]

Picks

• Andrew Mayne: Fantastic Four — Andrew clearly says it is his first pick and that he really liked it, though he also thought it was edited too tightly and wanted more breathing room.
• Brian Brushwood: Predator — Brian describes watching the original Predator with his daughter and says it was amazing and held up well, despite some dated effects and language.
• Justin Robert Young: Wednesday — Justin explicitly says 'my pick' and recommends the new season, describing it as a goth-Harry-Potter-like setup that works.
• Brian Brushwood: Alien: Earth — Brian says he watched the first half of the series and loved it, although he acknowledges it may feel slow to some viewers.
• Justin Robert Young: Alien: Earth — Justin says he watched the whole first episode and loved it, praising Noah Hawley and the setup.