9 ms·
I was wrong. Wrong for doubting the progress of AI, the pace at which things would improve, the usefulness of it all. I don’t subscribe to the idea of these mo
by pocketarc 4y ago
I was wrong. Wrong for doubting the progress of AI, the pace at which things would improve, the usefulness of it all.
I don’t subscribe to the idea of these models having any kind of intelligence (they’re just really good prediction machines), but man, they are getting unbeatable at their tasks, at a breakneck pace.
- Kim_Bruning 4y agoAn F-35 is just a really good kite, and a Bugatti Veyron is just a really good horseless carriage. And human intelligence is just a really good prediction machine. Entirely true, but somewhat misleading. The same goes for your statement "these models are just really good prediction machines". Seems a bit misleading. Mind you, qua terminology: there's a lot of levels of intelligence below human level (eg. insect level, or bird level, or etc...), and there's a lot of kinds of intelligence besides human (eg again: birds, insects, octopedes, etc). So we might be able to get away with calling this a kind of intelligence anyway, without beating around the bush.
- fnordpiglet 4y agoI’m not sure I don’t believe we aren’t very good prediction machines ourselves. Don’t forget people have to actually study logic, and studying is reinforcement of likelihood. And even with having practiced logic we make logical errors all the time despite feeling confident in our statements. I’ve found little in my life that tells me human intelligence isn’t just predicting the next “right” thought given experiences of should be the next right thought in a specific situation. Our “sense” or feeling of rationality, understanding, etc, may just be guides for probability of “right.” Confusion happens when a situation is novel enough that we can’t predict the next right thought. Etc. So, I’m actually not convinced what happens as the models produce an expected next token it’s not a form of intelligence in the same vein as human intelligence. I suspect as these models are pulled into more complex ensembles it’ll become striking how “intelligent” they are, and as they become less simply prediction engines how alien they becomes - because it would turn out we were just prediction models to begin with.