In this episode, we analyze Google’s latest power move with the upgraded Gemini 3 Deep Think and the Aletheia agent, which are currently dominating benchmarks in math, science, and autonomous research. We contrast this with OpenAI’s pivot to specialized hardware, breaking down the launch of the ultra-fast GPT-5.3-Codex-Spark on Cerebras chips.
The discussion also covers the new standard in AI video generation set by ByteDance’s Seedance 2.0—which has finally conquered the infamous "Will Smith eating spaghetti" problem—and the economic disruption caused by MiniMax’s low-cost coding models. Finally, we look at the "Pokémon Paradox": why frontier models can solve PhD-level physics problems but still struggle to beat a game on the Game Boy.
Topics Covered:
• Google's Deep Think: Crushing benchmarks and the rise of the Aletheia math agent.
• OpenAI x Cerebras: The move away from Nvidia and the release of the high-speed Codex-Spark.
• ByteDance's Seedance 2.0: Crossing the uncanny valley in video generation.
• MiniMax M2.5: How Chinese labs are driving down the cost of intelligence.
• The Pokémon Challenge: Why long-horizon planning in video games remains a hurdle for AI.
