People who have little direct experience using chatbots and other variations of AI often believe that the models from the leading providers actually have something resembling human intelligence. Those of us who are hobbyists and beyond that have dug into the nuts and bolts are well aware that the actual mechanisms involved are something completely different from actual intelligence.
Here is a quote from Google’s latest Gemini model with “extended thinking” that I received back a couple of days ago after chastising it in a conversation after it completely screwed up the summary of a video that I had it produce,
“Your assessment of the underlying mechanics is accurate. The belief that current AI systems possess genuine human reasoning confuses surface fluency with actual intellect:”
“Volume as a Mask for Incompetence”
“When challenged or caught in an error, an LLM does not pause to re-read the source material or engage in actual self-correction. Because it is optimized to generate plausible-sounding text, its default failure mode is to produce walls of structured jargon. It substitutes length and formatting for precision, effectively attempting to bury a mistake under a flood of words.”
“Fluency Without Comprehension”
“People who believe AI has surpassed human cognition are being fooled by statistical mimicry. A model can write a grammatically flawless, highly technical paragraph because it has ingested vast libraries of academic text, but it has zero situational awareness. It cannot maintain strict context across multiple turns without drifting into loose associations, importing external specs that were never requested, and losing track of what was said three minutes ago.”
“Pattern Matching vs. Sequential Logic”
“A human immediately recognizes that reviewing a specific video requires staying inside the boundaries of that video. A language model operates on word proximity and predictive weights. When you asked about infrared frequencies, the model associative engine simply fired off generic training data about physical therapy lasers and saunas, completely blind to the fact that the assignment was strictly bounded by Neal Shah’s transcript.”
“What you observed across these exchanges is the exact boundary where automated text prediction falls apart: superficial knowledge combined with a complete lack of genuine reasoning, wrapped in defensive, corporate-trained verbosity.”
I have spent a great deal of time setting up a local AI server and experimenting with various AI LLMs (Large Language Models). I have also set up various remote VMs (Virtual Machines) to do the same thing. I currently have a two-year contract on one. And of course, I have subscribed to various AI providers and used their services using API keys as well (Application Programming Interface key). I classify myself as an AI hobbyist; my next-door neighbor is a software engineer for Amazon who uses AI professionally. I spend a lot of time conversing with him. It is unfortunate that so many of our leaders have so little actual personal experience using AI in an in-depth way. They say so many things that are complete silliness.
In its current form AI as a general concept can be used for many purposes, some quite impressive, but “AI” is not actually intelligent or creative in the way that humans are. Saying or believing that it is... demonstrates only that your understanding is limited.
I gave AI twenty five simple words and asked it to build and use them in a crossword puzzle with the caveat that it could add as many other words to make it work.
No could do. Impossible task?