Darren Acemoglu's "5% job replacement" claim sparks TBPN debate on AI diffusion

Acemoglu's dual message: AI could lift growth, but adoption remains slow

The hosts discuss an essay by MIT economist and Nobel laureate Darren Acemoglu, summarized by Mustafa Suleyman, arguing that AI companies should "stop building AI to replace people and start building it to make them better at their jobs." Acemoglu's piece states that 52% of Americans are worried about AI's impact on their jobs, which he calls overblown fear that is nonetheless steering how AI is built. He predicts only 5% of human work will be replaced by AI in ten years and that AI will add roughly 1.5% to GDP over a decade, calling this "not a revolution."

One host pushes back using the example of Ridge, a ten-year-old company growing 50% this year using AI across its business, arguing that unlike electricity, AI can spread fast because it rides on existing internet infrastructure. Acemoglu anticipates this objection in his own essay, instead comparing AI to speech recognition: a technology pursued since the 1930s that saw a real breakthrough with Dragon Systems in the 1980s-90s, reaching "close to 95% accuracy" by 1997, yet consumer voice software today is "only marginally better" than in 2000 because of a string of corporate disasters (Dragon's acquirer going bankrupt, later owners deprioritizing consumers). Acemoglu uses this to argue that breakthroughs in AI infrastructure need the right applications, marketing, and consumer feedback loops to actually diffuse — conditions he says are not yet in place for AI.

A possible contradiction: Acemoglu on 60 Minutes says "No" to more jobs

The hosts flag an apparent tension between the essay and Acemoglu's appearance on 60 Minutes, where he is told "literally all of the top 20 economists in America" believe there will be more jobs in ten years than today, and asked "You're a top economist in America. What say you?" He replies simply, "No." One host argues the two positions are not actually incompatible: Acemoglu's point is that if AI continues to be built the way it currently is, it will replace workers, but if redirected toward empowering workers it could raise both employment and growth. The hosts note this reconciliation is their own interpretation rather than something Acemoglu states directly in the clips shown.

Mistral's "Le Chonk" and the open-weight AI race

Mistral launched Mistral Large 4, nicknamed "Le Chonk," described as a trillion-parameter, natively multimodal model with 49 billion active parameters. The hosts relay Mistral's own claims that it is the best open-weight model from the US or Europe on aggregated benchmarks and state-of-the-art on workloads including cyber defense, manufacturing, and finance, with open weights for a version planned for release "end of October." A host named Tyler says it looks "very much on par with the frontier of the open source" Chinese labs, calling it "certainly... a contender" for users wary of legal risk from Chinese open-source models, though not necessarily leading every benchmark.

A viewer comment disputes this, saying both Mistral's model and Reflection's new "Beam" model are "a generation behind the latest Chinese leading model." Tyler concedes the point, saying "it's not leading them, it's like among one of the contenders... not the frontier model for sure," while noting Chinese open-source releases happened only a couple of months earlier, shortening the usual catch-up window. Separately, the hosts mention Harvey's legal-agent benchmark, on which Mistral Large 4 leads at 15% versus other models scoring between 5-13%, though a host says he has "no idea" whether this reflects intentional optimization.

Consumer AI rankings and other notes

Citing data from Olivia Moore/a16z across web, mobile, and consumer revenue, the hosts report ChatGPT running at roughly 2x Gemini's scale and 6x scale on web, and about 2.5x Gemini and 14x Claude's scale on mobile. One host highlights Claude's trajectory in app store rankings, which has repeatedly risen to the top and then fallen as ChatGPT retakes the lead, describing it as "the war for the frontier... trading shots back and forth."

Separately, the hosts discuss a report describing Warren Buffett's nightly YouTube habits — watching old commercials, Paul Anka videos, and an "Uzbekistan's Got Talent" clip — relayed by his friend Sharon Osberg, with one host joking Buffett is "as locked in as ever" studying "the blade." They also debate a new US passport redesign, criticizing its typography and wide letter-spacing as resembling "AI slop," and discuss AI-driven video game decompilation projects, noting Ben Thompson's analysis that decompiled game mods likely violate copyright once shipped, even though disassembly itself can be fair use under precedents like Sega v. Accolade.

"He's saying that right now if we keep building AI the same way it will replace workers but we need to build it a different way that empowers workers." — a host