Last week the BBC sat down with with Yann LeCun.
If that name means nothing to you, here’s the plain version. He spent ten years as the chief AI scientist at Meta — the company that owns Facebook. He is one of a handful of people on this earth who can honestly say they built the field. When the history books get written, his name is in them.
And here is what he told the BBC, this month, July of 2026, about the AI systems the whole world is using right now — he named them: ChatGPT, Claude, Gemini.
He said they’re “not particularly smart.”
He said they accumulate knowledge and hand it back. They don’t understand what they’re saying. When one of these systems answers you, it isn’t reasoning about the real world. It is generating what looks statistically likely to be a good answer. Looks likely. Not is true. Those are two different animals, and the man who helped build the barn just told you which one is standing in it.
He gave a simple picture. Hold a pen upright on its tip and let go. A toddler knows it falls. No human wastes a breath guessing which direction, because there’s no way to know. But an AI will guess. It will produce a confident prediction, because producing confident-sounding output is what it was built to do. The confidence is the product. The truth is optional.
Now, some folks will read that and hear doom. I read it and heard confirmation.
Because a billion dollars heard it too. LeCun left Meta and started a new lab to fix the problem, and investors — including the chip giant Nvidia and the fund that manages Jeff Bezos’s own money — handed him one of the biggest early checks in European history. Serious money does not chase problems that aren’t real. The people closest to this technology just paid a billion dollars to agree that the machine on your desk generates plausibility, not understanding.
Here is where the kitchen table comes in.
Thirteen months of this operation’s daily record — and one protocol in particular — stand on exactly the diagnosis LeCun just published. On June 21, 2026, this operation ratified a protocol called POVL-1. Plain version: the machine’s natural pull is toward the answer that sounds right, so the checking has to happen before the answer forms, not after it’s already dressed up and served. That is the entire protocol, built from eighteen months of sitting with these systems daily and watching the confident wrong answer walk out the door wearing a suit.
Two weeks ago, the fourteen-rule public card went up on this site. Read rule two again: check before you claim — if you don’t know, say you don’t know. Read rule five: live facts get pulled live and named as pulled, never guessed. Those rules exist because of the exact behavior LeCun just described to the BBC. The pen falls. The machine guesses. The card says: don’t let it.
Now notice the two roads out of this, because this is the part that matters.
LeCun’s road is to rebuild the machine. New architecture, new math, a system that actually models the world instead of guessing at it. He says industrial trials next year, and real general intelligence somewhere down the line. It’s honest work and I hope he gets there. But down the line is not an address. You can’t sit at it this morning.
The other road is governance. You take the machine that exists today — the plausibility engine, his diagnosis, not mine — and you put conduct rules around it. You make it check before it claims. You make it name what it doesn’t know. You make it pull live facts live. You make the record the referee. That road is open right now, costs nothing to walk, and fits on one card a tenth grader can read in two minutes.
These aren’t competing answers. His is the engineering track. This is the governance track. Same admitted problem — the builders’ side just admitted it out loud. One fix arrives someday. The other one you can run before supper.
A builder knows this pattern in his bones. Nobody waits for the invention of a saw that can’t cut a man before they teach the crew to keep their fingers clear of the one on the bench. The safer saw is worth building. The safety rule is what keeps your hand whole while they build it.
So take the good news in this, because it’s genuinely good. The most credentialed man in the field just told the whole world the truth this operation has been publishing daily for over a year: the machine is a talker, not a thinker — yet. The fix is funded and underway. And the conduct standard that holds the floor until the fix arrives is already written, already public, already free, fourteen rules long, sitting one post back on this site.
The pen falls. Now you know why the card exists.
Dated July 11, 2026. The record holds.
Post Library – Intelligent People Assume Nothing
“If this post helped you understand AI better, then send it to one person who sees it like you do. Word of mouth is the only algorithm nobody owns.”
Purchasing Page – Intelligent People Assume Nothing
Contact: micvicfaust@gmail.com
© 2026 The Faust Baseline LLC | All Rights Reserved
