I'm done. I'm f***ing done.
In a Nutshell
The speaker rejects the concept of AGI entirely, arguing it lacks clear definitions, real problem-solving value, and any path beyond current LLMs' fundamental limits with context and understanding. He dismisses hype around future progress as propaganda, insisting LLMs are statistical tricks that only succeed by fooling CEOs into replacing workers, not by achieving intelligence. From now on, he will judge the technology strictly on its present capabilities without extrapolation or benefit of the doubt.
These notes were generated by AI and may contain inaccuracies.
At in-laws house on Sunday with food and family including two children, the speaker sat in the corner swiping on phone after having the thought at 2:00 a.m. the night before about whether AGI is possible. Posted a members-only video confessing to entertaining the idea that AGI could be possible in theory and is an engineering problem solvable in a couple hundred years.
After reading about AGI and intelligence from the moment of waking until going to sleep, decided to reject the concept entirely. Dipped one foot into the cult of AGI by accident and is now rejecting it completely.
David Deutsch, father of quantum computing and author of The Beginning of Infinity, does not believe animals are intelligent or conscious. The speaker found this position flabbergasted while trying to determine if a Python script can be generally intelligent.
If arguments are being made that animals aren't intelligent, then conversations about whether computers are intelligent become meaningless. There are no definitions for any of the terms being claimed to exist or be possible.
AGI is characterized as materialist propaganda wet dream. The speaker spent 12 hours reading about the topic and became exhausted, with wife asking to put down phone and acknowledge her mother while focused on determining whether intelligence is possible.
It is not clear what problem AGI would solve. Questions raised about why AGI would be needed for coding tasks and what it would offer beyond current generation LLMs. For coding specifically, generality is not wanted but rather specialization.
Mid-level engineer who has worked in a specific codebase for 10 years writes better code than a newly hired PhD-level engineer unfamiliar with the codebase. Expertise comes from carrying the whole system in one's head for many years and specializing in that specific codebase.
Current LLMs fail at coding because they cannot hold context in their limited capacity. The required context includes not just the codebase itself but all intangibles, business context, politics, and interpersonal dynamics such as a designer quitting over button color changes.
LLMs are doomed to fail in codebases by definition. AGI would not solve the fundamental context problem, understanding problem, or the requirement of being engrossed in the problem space. An AI cannot soak itself in the universe of the problem being solved, understand, feel, or engross.
David Deutsch argues that because intelligence arrived in a package at once in the form of a human, it does not make sense to separate intelligence from consciousness and personhood.
From now on, will stop pontificating about the future and stop giving technology the benefit of the doubt, benefit of progress, and extrapolation. Will treat the present state of LLMs as the only fact that exists.
Will behave as though there is no such thing as progress, no such thing as the future, and no such thing as 12 to 18 months from now. There is only now. Current LLMs are characterized as a trick rather than AI, tricking people into believing they are intelligent.
The mechanism matters absolutely. References David Deutsch's discussion that the Turing test is not about passing AI through some glory hole and determining if the other person thinks it tastes good. If the Turing test can be passed by tricking people, then the test itself becomes pointless.
You can polish a lie forever and it never becomes the truth. Adding a billion more parameters creates a bigger lie. LLMs are an elaborate trick with no theory inside, only brute forcing statistics to trick people for long enough durations to force entire companies to adopt them as intelligent tools.
The real Turing test is whether the human on the other end is stupid enough to replace their entire labor pool with this statistical tool that operates inside the distribution demo and claims it could build anything. The bot does not have to be intelligent, only more convincing than the median CEO is skeptical.
Dismisses the argument that the human brain might just be dumb statistics as well.
Will never again extrapolate, never again speak of progress, and never again give this technology the benefit of the doubt. Will take the present and evaluate it on its own merit, just like everything else purchased. Everything else is characterized as fear, confusion, uncertainty, doubt, and propaganda.
Keep Mo Bitar in your library
Save the videos and channels worth coming back to, and find them again in one place.
