Gary N. Smith

_{Senior Fellow, Walter Bradley Center for Natural and Artificial Intelligence}

Gary N. Smith is the emeritus Fletcher Jones Professor of Economics at Pomona College. He is the author of more than 100 research papers and 20 books, including The AI Delusion, (Oxford University Press, 2018; translated into Korean, Vietnamese, Chinese for sale in Taiwan, and simplified Chinese for sale in mainland China), The 9 Pitfalls of Data Science, with Jay Cordes (Oxford University Press, 2019; winner of the 2020 PROSE Award for Popular Science and Popular Mathematics), The Phantom Pattern Problem: The Mirage of Big Data, with Jay Cordes (Oxford University Press, 2020), Distrust: Big Data, Data-Torturing, and the Assault on Science (Oxford University Press, 2023), and Standard Deviations: The truth about flawed statistics, AI and big data (Duckworth, 2024)

Archives

Computer Intelligence Versus Human Intelligence

_{Professor Joseph Weizenbaum created a chatbot he named ELIZA that conversed with users the way a psychotherapist might}_{Gary Smith
January 13, 2026}

Even though the users knew they were interacting with a computer, many were convinced that the program had human-like intelligence and emotions and they happily shared their deepest feelings and most closely held secrets.

Large Language Models (LLMs) Are Inherently Frail and Unreliable

_{From my recent tests, things have not improved much. Even an apparent improvement did not last long. See for yourself}_{Gary Smith
January 5, 2026}

Post-training might stabilize the responses to specific inquiries but then the bot can just go off the rails again, as my tests showed.

The Core Problem with Large Language Models

_{LLMs are inherently unreliable, which means that failures are not incidental, easily fixable glitches}_{Gary Smith
January 1, 2026}

The data deluge exponentially increases the number of coincidental, useless statistical patterns — so the probability of useful patterns approaches zero.

Is OpenAI Approaching the Valley of Death?

_{Overpromising is still a big problem; in any event, the fate of Netscape looms}_{Gary Smith
December 22, 2025}

OpenAI needs to show that ChatGPT is more than just the first publicly available LLM. It has not done that and maybe never will.

Home Ownership: Madness Over 30-Year Mortgages

_{It is tempting to put borrowing and investment decisions in separate mental buckets, but they are intimately related}_{Gary Smith
November 21, 2025}

When considering a mortgage, the correct comparison is not total payments but the return on the borrowed money versus the loan’s annual percentage rate.

Illusions: No, Large Language Models Do Not Understand

_{A recent New Yorker article is mistaken about this. For one thing, the LLMs have trouble distinguishing between causation and mere correlation}_{Gary Smith
November 14, 2025}

A real-world example of such struggles is the poor performance, in general, of AI-powered mutual funds.

Scaling Up is Not Going to Make Large Language Models Intelligent

_{ChatGPT-5 misunderstood the jab in a simple Will Rogers joke five out of six times}_{Gary Smith
October 30, 2025}

Bigger data centers are not going to change the harsh reality that LLMs are prone to generating unreliable blather.

Large Language Models: Inconsistent Because They’re Unintelligent

_{Here’s what happened when I tested popular LLMs on student exercise questions I have been using for over fifty years}_{Gary Smith
October 28, 2025}

Claims that an LLM is intelligent don’t square with the fact that it may answer a question one way and, a few seconds later, simply contradict itself.

For a Sounder Approach to Investment: Deep-Six the 60/40 Rule

_{Sometimes, good decisions require more than blindly accepting advice from humans as well as computers!}_{Gary Smith
October 20, 2025}

The conventional belief that a sound investment portfolio should be 60% stocks and 40% bonds does not stand up to careful scrutiny.

Intelligence Demands More than Pressing a Lever to Obtain Water

_{I continue to be astonished by how willing people are to assume that LLMs are intelligent because they give glib, confident answers}_{Gary Smith
October 7, 2025}

Human trainers can nudge LLMs to give good answers for specific questions but this training doesn’t prepare them for subtle variations.

Glued to Our Phones. Our Troubling Addiction to Social Media

_{It's disconcerting to see people out for a nature walk — completely absorbed in the traffic on their phones}_{Gary Smith
October 6, 2025}

Social media addictions clearly cause harm among minors but the tech giants who profit by them won’t regulate themselves in any meaningful way.

A Realistic Direction for Artificial General Intelligence Today

_{Based on GPT5's performance to date, it would make a superb substitute for a mansplainer I know of — call him Brock}_{Gary Smith
September 16, 2025}

The fundamental problem remains; models like GPT-5 are hobbled by their inherent inability to relate the text they input and output to the real world.

GPT 5.0 Doesn’t Understand But Is Eager to Please

_{Over a number of tries, it couldn’t get the labels on an illustration right because it does not understand what the words mean and how they relate to the image}_{Gary Smith
September 3, 2025}

The test also shows GPT 5.0’s inclination to praise a user’s acuity, whether the user’s comment is correct or incorrect, intelligent of dumb.

What Kind of a “PhD-level Expert” Is ChatGPT 5.0? I Tested It.

_{The responses to my three prompts made clear that GPT 5.0, far from being the expert that CEO Sam Altman claims, can’t address the meanings of words or concepts}_{Gary Smith
August 25, 2025}

In an era where politicians, celebrities, and businesses can get away with blatant untruths with little or no consequence, will the same be true of LLMs?

ChatGPT-5 Tries Out “Rotated” Tic-Tac-Toe. You Be the Judge…

_{They say that dogs tend to resemble their owners. Chat GPT very much resembles Sam Altman — always confident, often wrong}_{Gary Smith
August 22, 2025}

It’s no mystery why LLMs aren’t intelligent in any meaningful way. The real mystery is why so many otherwise intelligent people still take the claims seriously.

AI Is a Long Way From Replacing Software Coders

_{Despite C-suite claims, they are not likely to take our jobs any time soon because they do not understand what words mean and how words relate to the physical world}_{Jeffrey Funk and Gary Smith
August 4, 2025}

LLMs excel at simple coding tasks but are still too unreliable to use without extensive human supervision on complex tasks where mistakes are expensive.

Sci Foo Unconference: Horseshoe Crabs, Alchemy and (Of Course) AI

_{It’s called an “unconference” because attendees do not present papers in pre-organized sessions; they propose topics at the venue and those with most interest are selected}_{Gary Smith
July 22, 2025}

Many Sci Foo attendees were excited by LLMs in education but I fear that — when the way they are used is considered — they will increase social inequality.

The Job Market is Telling Us Something About AI and Jobs…

_{But it’s not telling us the same things as the AI hypesters are telling us}_{Jeffrey Funk and Gary Smith
July 1, 2025}

The job market for recent college grads may be warning us that participation trophies for college and ChatGPT use have long-run costs.

No, Large Reasoning Models Do Not Reason

_{Large reasoning models continue the Large Language Model detour away from artificial general intelligence}_{Gary Smith
June 13, 2025}

It is increasingly recognized that, without extensive post-training, their authoritative answers are often remarkably bad.

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.