Gary N. Smith

Gary N. Smith is the Fletcher Jones Professor of Economics at Pomona College. His research on stock market anomalies, statistical fallacies, the misuse of data, and the limitations of AI has been widely cited. He is the author of more than 100 research papers and 18 books, most recently, Standard Deviations: The truth about flawed statistics, AI and big data, Duckworth, 2024.

Archives

LLMs Are Bad at Good Things, Good at Bad Things

_{LLMs may well become smarter than humans in the near future but not because these chatbots are becoming more intelligent}_{Gary Smith
May 28, 2025}

As people become attached to and dependent on their AI friends, they become less interested in their fellow humans.

Yes, Large Language Models May Soon be Smarter than Humans…

_{But not for the reason you think}_{Gary Smith
May 9, 2025}

Too many students are not learning how to think and write; they are learning how to use LLMs — no matter that LLM responses cannot be trusted.

LLMs Still Cannot be Trusted for Financial Advice

_{The limitations of Large Language Models (chatbots) are illustrated by their struggles with financial advice}_{Gary Smith, Valentina Liberman, and Isaac Warshaw
April 21, 2025}

The real danger remains: we may think computers are smarter than us and consequently trust them to make decisions they should not be trusted to make.

Large Language Models: A Lack-of-Progress Report

_{They will not be as powerful as either hoped or feared}_{Gary Smith
April 14, 2025}

LLMs are enormously expensive yet it is increasingly clear that they will never be reliable enough to be trusted in complex situations where mistakes are costly

Machine Learning Algos Often Fail: They Focus on Data, Ignore Theory

_{Without a theory, a pattern is just a pattern}_{Gary Smith
April 8, 2025}

The data-mining algorithms may recognize that a statistical pattern has vanished but won’t know why because they do not understand it as a logical relationship.

Yes, the AI Stock Bubble Is a Bubble

_{It's unfolding the way a financial bubble typically does}_{Gary Smith
March 28, 2025}

Unlike the Internet, LLMs are not useful enough to prompt customers to pay prices that reflect the hundreds of billions of dollars needed to develop them.

Why LLMs Are Not Boosting Productivity

_{If LLMs were as reliably useful as economist Tyler Cowen alleges, businesses would be using them to generate profits faster than LLMs generate text. They aren’t.}_{Jeffrey Funk and Gary Smith
March 6, 2025}

So far, AI is dragging down economic growth by diverting so much human talent and natural resources away from more productive uses.

Intelligence Requires More Than Following Instructions

_{Post-training improves the accuracy and usefulness of LLMs but does not make them intelligent in any meaningful sense — as the Monty Hall problem shows}_{Gary Smith
March 3, 2025}

The danger is not that computers are smarter than us but that we think they are and thus trust them to make decisions they should not be trusted to make.

The Large Language Model (LLM) “Superpower” Illusion Dies Hard

_{Historic confirmation bias around ESP and spirit cabinets makes for an interesting comparison with the current need to believe in the abilities of LLMs}_{Gary Smith
February 24, 2025}

Someone blinded by confirmation bias might believe that robots and Go victories are evidence that ChatGPT would be a reliable lawyer.

Why LLMs (chatbots) Won’t Lead to Artificial General Intelligence

_{The biggest obstacle is seldom discussed: Most consequential real-world decisions involve uncertainty}_{Gary Smith
February 20, 2025}

LLMs can’t help when key decisions don’t feature objectively correct probabilities but rather subjective probabilities that need interpretation.

Some Lessons From DeepSeek, Compared With Other Chatbots

_{I tested OpenAI o1, Copilot, and Gemini Flash, along with DeepSeek, on a question about Tic-Tac-Toe}_{Gary Smith
January 30, 2025}

Here’s the problem with the use case for LLMs: If you know the answer, you don’t need to ask an LLM; if you don’t know the answer, you can’t trust one.

Sloppy Science is a Statistical Sin

_{Evidence of sloppy science encourages readers to wonder if the entire research project is compromised}_{Gary Smith
January 17, 2025}

Journals should require authors to make their data available online before publication and encourage others to try to replicate the research they publish.

The Hype and Limitations of Generative AI

_{Robert J. Marks and Gary Smith
January 16, 2025}

On this episode, host Robert J. Marks concludes his conversation with economics professor and author Gary Smith about the hype and limitations of generative AI. Smith is the Fletcher Jones Professor of Economics at Ponoma College and a frequent contributor to Mind Matters News. In this portion of the conversation, Smith and Marks explore the hype around artificial general intelligence (AGI) and explain how current large language models lack true reasoning and creative capabilities, despite regular claims of impending AGI from people like OpenAI’s Sam Altman. Smith provides examples demonstrating how these models make nonsensical or incorrect responses to logical problems and financial questions, highlighting their inability to understand context and perform meaningful reasoning. …