The AI Frontier: Deep Dive into DeepSeek, O3, and Beyond

Episode Notes

The episode opens with a long discussion of DeepSeek, its V3 and R1 reasoning models, and why the release caused such a big reaction in AI circles and on Wall Street. Andrew says DeepSeek appears to have made real efficiency gains in training and hardware use, while Justin argues the market overreacted to the idea that less compute would be needed; both stress that the models do not mean chips or compute are suddenly unnecessary (L17-L17, L25-L25, L49-L49, L53-L57, L61-L65, L73-L77, L101-L105).

The conversation then shifts to OpenAI's O3 and to a live, hands-on demo of generating simple games and 3D scenes with AI. Brian and Andrew iterate on a crude side-scrolling Mobius-strip game in CodePen, then experiment with A-Frame, a generated planetarium, and an explainer for radio telescopes, using the examples to argue that AI is becoming a practical tool for prototyping, brainstorming, and building educational or creative projects faster (L117-L117, L123-L145, L149-L181, L191-L209, L235-L241, L247-L253, L275-L289, L315-L317, L323-L333).

Key topics

• DeepSeek efficiency gains and model optimization: Andrew describes DeepSeek as having made real, original optimization improvements, especially around data movement, compression, and training efficiency under chip export constraints.
• Uncertainty about data provenance and bootstrapping: The hosts note possible data contamination or use of model outputs, but they are careful to say those suspicions do not fully explain DeepSeek's success and are not proven.
• Wall Street's reaction to AI compute narratives: Justin says the DeepSeek reaction hit Microsoft and NVIDIA hard and that investors briefly bought into a story that massive compute might no longer be required.
• Reasoning models like O1, O3, and R1: Andrew explains reasoning models as systems that work through subproblems on a scratchpad-like process before giving a final answer.
• Jevons paradox and expanding demand for compute: The hosts explicitly invoke Jevons paradox to argue that cheaper, more efficient compute tends to increase total usage and unlock new applications rather than reduce demand.
• Open source and democratized AI benefits: Justin argues that open source is central to AI adoption and points to a dentistry/x-ray example to show how cheap or free access can produce real-world benefits.
• CodePen, JavaScript, and prompt-driven game iteration: Brian and Andrew walk through taking generated code, moving it from Python to JavaScript, pasting it into CodePen, and iterating with prompts to make the game more interesting.
• Mobius-strip game concept: Brian's game concept explicitly uses a Mobius strip theme, with looping play and flipped controls.
• A-Frame and generated 3D environments: Andrew shows how to use A-Frame to generate and run simple HTML-based 3D scenes like a castle or planetarium.
• AI as a tool for business and project strategy: Brian uses GPT to analyze his audience and brainstorm a new live-stream/video-podcast project, and Andrew frames this as an example of AI helping with planning and positioning.
• AI for educational explainers: Andrew demonstrates generating a scrollable HTML explainer for radio telescopes, arguing that AI can lower the barrier to educational content creation.
• Prompting as management and iterative refinement: Andrew says prompt engineering is less about magic wording and more about giving clear, manager-like instructions and then refining outputs.
• Creative collaboration versus fully personalized output: The hosts suggest AI will enable much bigger creative projects, but Andrew says the best results will still likely be collaborative rather than totally individualized.

Picks

• Andrew Mayne: Severance — Clear recommendation; Andrew explicitly says 'my pick is right now, severance' and says he has been loving it.
• Brian Brushwood: Severance — Strong endorsement after discussion; Brian says he absolutely adores the show and then 'double[s] down on severance as well.'