Mistral's Le Chonk and Daron Acemoglu reveal the limits of AI
By Imani Sutton · Reporting from Atlanta ·
On TBPN, hosts debated AI economic impacts and Mistral's new model, ignoring the grid and water costs shouldered by everyday ratepayers.
On a recent episode of the tech talk show "TBPN", hosts John Coogan and Jordi Hays traded takes on artificial intelligence, Warren Buffett's screen habits, and Mistral's latest large language model release. The show kicked off with a discussion about a leaked six-hour instructional video on screen acting by Tom Cruz. From there, the conversation shifted to video game decompilation driven by large language models, where Ben Thompson's analysis of copyright implications and viral mashups took center stage. As host John Coogan noted, "Games are written in human readable languages, C, C++. Obviously they're scripting languages on top of that, but then they're compiled into machine code that's executable by the processor that intended for."<br><br>## The productivity paradox and the ghost of Dragon Systems<br><br>Later in the episode, the discussion turned to the actual economic impact of artificial intelligence, featuring MIT economist Daron Acemoglu's cautious views on diffusion and job displacement. Acemoglu compared the slow spread of AI to the century-long quest for speech recognition, highlighting the cautionary tale of Dragon Systems, founded by James and Janet Baker in the 1980s. As recounted on the show, Dragon's revolutionary speech-to-text software was repeatedly derailed by corporate disasters, including an acquisition by Learnout and Housepie that ended in bankruptcy and fraud convictions, followed by Nuance and eventually Microsoft prioritizing enterprise and medical markets over consumers. Coogan observed that "today breakthroughs in the infrastructure of a new technology need to go handinhand with the right kinds of applications built on this infrastructure." Yet, the structural reality remains that massive compute centers demand enormous power and water resources while corporate actors capture the margins.<br><br>## The algorithmic rabbit hole and French model weights<br><br>The episode also touched on lighter fare, including a Wall Street Journal report on Warren Buffett's nightly YouTube rabbit holes, watched from his Omaha home with friend Sharon Osberg, featuring everything from 1950s commercials to the Uzbek version of America's Got Talent. Toward the end, the hosts covered Mistral's launch of their new model, dubbed "Le Chonk", featuring one trillion parameters and positioning itself as a top open-weight competitor. Whether these models are deployed via Mistral's own cloud infrastructure or run locally, every token generated still traces straight back to a data center drawing megabawatts from local utility grids and pushing the environmental and financial costs onto ratepayers. We do not need more press-release futurism about algorithms; we need accountability for the infrastructure that powers them, billed directly to the companies that profit from our scarcity.