“The Big Short” investor Michael Burry is warning against industry spending of about $1 trillion on artificial intelligence, comparing the focus on large language models to a 19th-century bubble.
Using an 1880 newspaper case study, Burry argues that tech capital expenditure is misdirected, favoring the “System 2” architectures from Meta Platforms Inc. (NASDAQ:META) and Google’s parent Alphabet Inc. (NASDAQ:GOOG) (NASDAQ:GOOGL) over text-predicting models.
Thought Before Language
Burry based his critique on an 1880 New York Times edition detailing a Smithsonian Institution case study titled “Thought Without Language.” The study observed Melville Ballard, a deaf-mute child who developed complex existential inquiries regarding the origins of the universe before acquiring written or sign language.
The account described Ballard communicating abstract ideas through gestures and expressions before acquiring conventional language, which Burry uses as evidence for his argument that reasoning can precede language.
Applying this physiological reality to artificial intelligence, Burry asserts that large language models are “built backward” because they attempt to scale language into reason. This method results in a primitive simulation highly prone to hallucination. According to Burry, language should be the output of understanding, not its core engine.
A ‘Parameter Trap’ and Speculative Gluttony
Pointing to a second article on the same 1880 newspaper page about San Francisco “bonanza speculators” who triggered absurd market valuations in a quest for immediate fortune, Burry draws a parallel to Silicon Valley’s AI hardware frenzy.
He characterizes the current market environment as a “parameter trap” defined by “supply-side gluttony and massive capital misallocation.”
Burry warns against the “singularity” myth of scaling models in an “infinite mirror.” He views this trajectory as a directionless waste of computing resources that is completely disconnected from economic reality.
The Shift to ‘System 2’ Reasoning
Rather than abandoning AI, Burry points to reasoning-oriented approaches as a potential alternative to scaling conventional language models.
Burry points instead toward reasoning-oriented AI research, highlighting Google’s AlphaGeometry and Meta’s Coconut. AlphaGeometry combines neural and symbolic reasoning, while Coconut explores reasoning in a continuous latent space rather than relying entirely on natural-language intermediate steps.
These architectures emphasize compression and logic over brute-force text prediction, aligning with the 145-year-old revelation that true comprehension transcends language entirely.
How Have These Stocks Performed?
| Stocks | 1-Month | 6-Months | YTD | 1-Year | 5-Years |
| META | 25.46% | 35.2% | 9.86% | -1.25% | 113.67% |
| GOOGL | -0.72% | 25.81% | 9.93% | 41.54% | 157.4% |
Disclaimer: This content was partially produced with the help of AI tools and was reviewed and published by Benzinga editors.
Photo courtesy: Shutterstock
Login to comment