David Manheim contrasted modern progress with Isaac Asimov's predictions.
When I was a kid I didn't realize, as is now obvious, that AGI would be a wide band rather than a sharp finish line. But if you'd shown me a version of ChatGPT that had been told to act like a human, I'd definitely have said that AI had been achieved.
Many users praised Paul Graham's framing of human-like ChatGPT as marking real AI achievement and called it insightful or already reached, while others objected that goalposts keep moving and development prioritizes ads over meaningful AGI.
@paulg i was just wondering about this and i think AGI would not be possible with the current architecture of LLMs and i know that many people are saying this and i din't discover it myself but i am just starting to believe in this notion more, too many bottlenecks in the process to AGI
@paulg Yet we are like animals born in a zoo, suddenly released into the jungle, still mostly pacing in our invisible cages. The biggest unlock isn't better models; it's unlearning our old constraints.
I wasn't worried about it. I was excited about it. Reading The Moon Is a Harsh Mistress and seeing SHRDLU on TV was what made me want to work on AI as an undergrad.
@paulg When I was a kid, the idea that any technological transformation would occur in less than a decade was fanciful. Asimov's timelines for the widespread adoption of positronic robots spanned multiple centuries!
@matterasmachine @paulg ¿Cómo decide una red neuronal natural qué conexiones podar y qué conexiones conservar?🤔
@paulg Every big tech company is racing to build AI, smarter shopping, better ads, more code, more images, more videos… but it feels like they’re missing the point. Most people don’t need another AI feature rn, they need more money in their bank account.
@paulg AGI will emerge from decentralized networks like Bittensor and not from any centralized AI labs.
@paulg This is so fun to read from the author of the book we used in my first AI class at GA Tech, CS 20something - Knowledge Representation and Processing
@matterasmachine @paulg Es de cajón, y matemáticamente demostrable. La capacidad de decisión de la inteligencia natural como mínimo duplicará a la inteligencia artificial. La AGI no es más que una maqueta aproximada de la inteligencia natural, solo eso.😉
@paulg It should not be what AGI can do. It should be what a human can do. A human can invent and discover - proven by history. AGI can mix data.
@paulg Oddly when I was a kid in the 1970s, it was quite widely believed by people who knew nothing at all about computers and had never seen one (ie most people) that computers kinda knew everything, could answer difficult questions, etc
@paulg AGI will feel like the horizon. The models keep improving, we get used to them in a week and ask for improvements, and the definition moves further out each time.
@paulg you shouldn't have worried about AGI when you were a kid paul
it's all a wide band It's shocking how things look in retrospect Doesn't it feel like history turned a page in these last 3 or so years? Maybe 6 but it was unignorable in the last 2.
@paulg It depends on how fast AI progresses. If it progresses fast enough — especially if the rate of progress continually increases — the wide band could fairly suddenly turn into a sharp finish line.
I don’t think it’s obvious that this won’t happen.
@paulg A wide band makes sense; a model that could convincingly keep up the human act would have cleared most childhood definitions of AI.
@paulg yeah but i think there might still be some psychological hurdles for us to overcome in order to fully embrace them as our equals
@matterasmachine @paulg Es, sin duda, una decisión que tiene que ver con el gradiente.
@paulg October 2025 felt like a pretty sharp line to me
@matterasmachine @paulg La inteligencia natural cuenta con un "testigo". Este testigo se encarga de registrarlo todo sin evaluarlo. Cuando la AGI cuente con este tipo de elemento ... seguiremos hablando ...
David Manheim contrasted modern progress with Isaac Asimov's predictions.
When I was a kid I didn't realize, as is now obvious, that AGI would be a wide band rather than a sharp finish line. But if you'd shown me a version of ChatGPT that had been told to act like a human, I'd definitely have said that AI had been achieved.