AI-Lawyers Will Have Fools for Clients
When tech prophets promise everything—except accuracy.Lazy lawyers have no doubt filed hallucinated content orders of magnitude more often.
Read More ›
Gary Smith
Computer Intelligence Versus Human Intelligence
Professor Joseph Weizenbaum created a chatbot he named ELIZA that conversed with users the way a psychotherapist mightEven 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.
Read More ›
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 yourselfPost-training might stabilize the responses to specific inquiries but then the bot can just go off the rails again, as my tests showed.
Read More ›
The Core Problem with Large Language Models
LLMs are inherently unreliable, which means that failures are not incidental, easily fixable glitchesThe data deluge exponentially increases the number of coincidental, useless statistical patterns — so the probability of useful patterns approaches zero.
Read More ›
Is OpenAI Approaching the Valley of Death?
Overpromising is still a big problem; in any event, the fate of Netscape loomsOpenAI needs to show that ChatGPT is more than just the first publicly available LLM. It has not done that and maybe never will.
Read More ›
Home Ownership: Madness Over 30-Year Mortgages
It is tempting to put borrowing and investment decisions in separate mental buckets, but they are intimately relatedWhen considering a mortgage, the correct comparison is not total payments but the return on the borrowed money versus the loan’s annual percentage rate.
Read More ›
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 correlationA real-world example of such struggles is the poor performance, in general, of AI-powered mutual funds.
Read More ›
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 timesBigger data centers are not going to change the harsh reality that LLMs are prone to generating unreliable blather.
Read More ›
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 yearsClaims 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.
Read More ›
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!The conventional belief that a sound investment portfolio should be 60% stocks and 40% bonds does not stand up to careful scrutiny.
Read More ›
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 answersHuman trainers can nudge LLMs to give good answers for specific questions but this training doesn’t prepare them for subtle variations.
Read More ›
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 phonesSocial media addictions clearly cause harm among minors but the tech giants who profit by them won’t regulate themselves in any meaningful way.
Read More ›
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 BrockThe 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.
Read More ›
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 imageThe 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.
Read More ›
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 conceptsIn 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?
Read More ›
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 wrongIt’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.
Read More ›
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 worldLLMs excel at simple coding tasks but are still too unreliable to use without extensive human supervision on complex tasks where mistakes are expensive.
Read More ›
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 selectedMany 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.
Read More ›
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 usThe job market for recent college grads may be warning us that participation trophies for college and ChatGPT use have long-run costs.
Read More ›
No, Large Reasoning Models Do Not Reason
Large reasoning models continue the Large Language Model detour away from artificial general intelligenceIt is increasingly recognized that, without extensive post-training, their authoritative answers are often remarkably bad.
Read More ›