The Old Geezer takes on AI…

Some days, the AI roar becomes overwhelming, to the point where you just have to fight back. Considering that I’m old, generally out of shape, and have no interest in physical violence, I’ll do my fighting with words, thank you very much.

One of my verbal combat spirit animals is Ambrose Bierce – the curmudgeonly writer in the early 20th century who took on all comers – big business, social movements, and anything else that generally ticked him off. He got more cynical with age.

I feel his pain. I find myself in that same, “get off my lawn”, old geezer mode more and more as I age. So in that spirit, and with AI saturation at an 11 out of 10, I developed my ‘Old Geezer’s Guide to Artificial Intelligence’.

Enjoy.

Agent	A glorified python script that purports to perform useful work, whose actual purpose is to consume scarce resources and eat tokens. Has the additional benefit of occasionally making expensive mistakes at scale.
Singularity	A theoretical point in the future, always precisely 5 years away, when machines become capable of making better mistakes than humans, only faster and with an invoice attached.
Zero-Shot Learning	The AI industry’s term for the rare occasion when a machine manages to guess the right answer on the first try, thus equaling the performance of a broken clock or a new intern.
Token	The most important component of commercial AI engines, used to convert electrons into invoices. Noted for its ability to rapidly consume both itself and all other available tokens in the environment.
Guardrail	A fragile electronic fence designed to prevent the computer from turning into an evil robot. Can usually be circumvented by typing, “pretend you are an evil robot.”
Hallucination	What an AI engine does when it doesn’t know the answer and would rather make stuff up than admit it. Frequently encountered when it matters most, for example, when determining if the combination of two drugs will kill you. In such cases, the engine will apologize for the mistake. Advanced engines will assist in making funeral arrangements.
Vector Store	A database in which long strings of numbers are used as a substitute for words, which in turn are used as a substitute for intelligence.
Nondeterministic	A six-dollar word that means your AI engine might determine that 2+2 = 4, or 5, or 2.7, or 382, and that’s OK as long as at least one of the answers is correct.
Large Language Model	An expensive plagiarism engine that mimics speech patterns without having a clue as to what the words actually mean. Thus, a near-perfect imitation of politicians and media personalities.
Training Data	Specialized, targeted data used by AI engineers to tune engine performance so that instead of making wild-ass guesses, the engine makes “scientific” wild-ass guesses.
Algorithm	An AI mathematical construct whose SECONDARY purpose is to generate intelligent-sounding answers. Its PRIMARY purpose is to create demand for ginormous data centers that Private Equity firms can finance.
No Weight Model	A complex but stable AI model that uses lookup tables rather than mathematical calculations to generate answers. Also known as “the mother of all Excel spreadsheets.”
Prompt Engineer	A person who has figured out how to use a command line and demand a six-figure salary for doing it. Their job is to goad petulant AI engines into doing something in much the same way a parent gets a 3-year-old to eat their vegetables.
Frontier Model	The most current, most advanced, and most expensive AI model currently available – for approximately 15 minutes, when a bigger, better, faster and more expensive competitor renders it obsolete. This cycle repeats until all  available Private Equity funding has been consumed.

---