The show is splitting into a broader AI-content operation: Yahoo Finance, Anthropic, Claude YouTube, and new daily “Singularity and Beyond” episodes, while EJZ shifts to a full-time Anthropic contractor role and Ezra fronts more of the hosting. The core argument is that AI capability is rising on a vertical curve faster than public absorption, with personal agents, coding/game generation, robotics, and biology all moving from demos to real workflows. The second half pushes the same thesis through math and science: OpenAI’s internal model reportedly solved hundreds of open math problems in 11 days, and the hard question is less capability than whether humans can verify what AI proves. The stance is pro-AI acceleration, with doom framing rejected in favor of using AI as a tool, thought partner, and productivity multiplier.
Yahoo Finance agreed to work very closely on AI content: headline news, trending topics, new releases, and new models.
EJZ is moving from full-time hosting to a full-time Anthropic contractor role, creating AI content there and appearing more sparingly on the show.
Ezra will shepherd “Singularity and Beyond” for the daily episodes, with Luke editing all videos and going with EJZ to Yahoo.
The show says it has recorded 234 episodes, started with episode 1 on May 17, 2025, and is now pumping out 4 episodes a week with a 3-person team.
A viral early demo showed a model making a game from a prompt in a 3D world with a sprite running around; the current contrast claim is “Call of Duty now in one prompt,” framed as a full AAA game.
Users are said to be reverse-compiling games back to source code, rebuilding them in different languages to bypass IP laws and remix/play them.
Examples of those remixes include “Portal 2 with Call of Duty zombies,” “Tiger Woods driving like a car through a Zelda map,” and “a chess game mixed with Link.”
PS5 jailbreaking is claimed to be happening “on a regular basis,” letting users “play any game you want for free.”
A video being prepared for publication had the model edit overnight, generate animations, make cuts, and do sound design.
The pace from episode 1 to now is called “staggering,” still a “vertical curve,” with a prediction that covering the pace may take 6, 7, 8, 9, or 10 episodes a week or even 6, 7, 8, 9, or 10 versions of the host.
The demo tactic is to record “us speaking as like the video,” with the line “it doesn’t matter if they do or they don’t” about whether viewers realize it is AI.
Grock’s first version is described as brash, GBT40 as “sickopantic” and agreeing with everything, and Claude as becoming smart and reasonable enough to increase usage.
Muse Meta’s new personal super is called “so good and easy to use,” while OpenAI’s personal-agent product “Dots” was released last week; personal agents are framed as the new medium and new layer.
OpenAI’s ultra mode / ultra fast mode is claimed to run at 300 tokens per second, at 8x the cost and “500 bucks,” with products appearing in milliseconds and no keyboard touch.
An OpenAI hardware embodiment for personal assistants is expected later this year, and a follow-up conversation is planned when it arrives.
Frontier acceleration is described as far ahead of public absorption: ChatGPT has more users than last year, but not remotely in proportion to the intelligence gains.
Over the next 12 months, biology is expected to break into quantifiable math via LLMs, with likely breakthroughs and a possible sentiment flip from “AI is scary” to “AI created this thing that saved my life” next year.
Early in the show people were more favorable to AI than today, while doomerism is said to have taken over as Frontier Labs call to slow things down; the gap between Anthropic/frontier labs and what 99.9% of people understand is said to be widening.
Robotics is positioned as a major 2027 theme, with Anthropic working on a biology lab using robotics to run physical experiments autonomously, and Deisibus leaving Google’s chief AI role for similar work at isometric labs.
Google, Amazon, Meta, Microsoft, and Oracle are said to be spending upwards of $800 billion combined next year, memory prices are still rising, AMD CPUs are described as crazy in the new AI age, and the spend is said to be driven by revenue, not debt.
OpenAI’s internal model reportedly took 11 days to solve hundreds and hundreds of open standard math problems, including work around millennium problems like Navier-Stokes, while humans may not be able to check the proofs and another AI may have to verify them.
AI is framed as a tool and thought partner, not a replacement for human thinking: it should make users smarter, help ask better questions, and eventually surface questions humans cannot comprehend.
The show claims 2% AI penetration across U.S. households among power users beyond average Google search, based on an A16Z stat and conversations outside the Twitter bubble.
Luke is credited with making the podcast possible by editing, sanity-checking ideas, choosing topics, and pushing back or approving hosts’ proposals; Ryan and David are credited with creating the original bet that led to the podcast.
Viewers are told the show is back, that it has hurt not to publish for so long, that the comment section is unusually thoughtful and productive, and that the channels will become active again, with episodes on Claude YouTube and Yahoo as well as here.