“ Simply retrieving a reasoning trace looks a lot like human reasoning, until it's time to navigate uncharted territory. If you memorized all reasoning traces of humans from 10,000 BC, you could automate their lives but you could not invent modern civilization.”
“Simply retrieving a reasoning trace looks a lot like human reasoning, until it's time to navigate uncharted territory. If you memorized all reasoning traces of humans from 10,000 BC, you could automate their lives but you could not invent modern civilization.”
I have. It's very nice for generating one-time tools, generating diagrams, explaining existing source code and refactoring.
Coding new functionality is ok at times, but it makes decisions that would be catastrophic once the app is scaled. Very nice for proof of concepts, but only somewhat helpful for production code.
I also think they are overselling it, even though I agree it's not just an autocomplete.
Or free advertising for an advanced autocomplete model?
edit:
Dario has been saying these lies since GPT2.0 in 2019.
He uses it as marketing.
He reverse engineered some research in 2025 saying “LLMs will blackmail users to keeping them switched on”
of course.
also telling that they don't give a shit about election fraud anymore.
>advanced autocomplete
So you haven't used claude at all then?
It’s not creating its own knowledge. Only mining human code
So you haven't used it then?
Are you trying to get tased?
“ Simply retrieving a reasoning trace looks a lot like human reasoning, until it's time to navigate uncharted territory. If you memorized all reasoning traces of humans from 10,000 BC, you could automate their lives but you could not invent modern civilization.”
Nice goalpost shifting.
And you clearly haven't used it for anything worthwhile if you're still desperately clinging to "autocomplete model."
“Simply retrieving a reasoning trace looks a lot like human reasoning, until it's time to navigate uncharted territory. If you memorized all reasoning traces of humans from 10,000 BC, you could automate their lives but you could not invent modern civilization.”
I have. It's very nice for generating one-time tools, generating diagrams, explaining existing source code and refactoring.
Coding new functionality is ok at times, but it makes decisions that would be catastrophic once the app is scaled. Very nice for proof of concepts, but only somewhat helpful for production code.
I also think they are overselling it, even though I agree it's not just an autocomplete.