12 ms·
Yep, I was just about to type pretty much the same thing. AI is wrong because, much like us, it's trained on contradictory and occasionally incorrect informati
by smoldesu 4y ago
Yep, I was just about to type pretty much the same thing.
AI is wrong because, much like us, it's trained on contradictory and occasionally incorrect information. To fix that using traditional methods, you'd need to train an AI on non-contradictory material that is objectively true. Training a model needs more than math textbooks and news articles though, otherwise it's not very generally-intelligent.
In the long-run, it's possible that AI of ChatGPT's scale will never be "correct" because it doesn't have a concept of correctness in the first place. It just generates text, and if a human deems that it fails then it generates it again. "Fixing" the training data limits the model's understanding, which leaves us with unreliable preprocessing/postprocessing methods.
- logicallee 4y agoPerhaps there could be a standard of "general intelligence" that doesn't require it to be right all the time, but merely have the ability to reason based on its training data and conversation in plausible ways? Currently, when it fails, it does so in a way that people don't. How long might it take large AI models to reach this different standard of general intelligence, requiring them to fail in plausible ways rather than completely showing a lack of reasoning and understanding?
- smoldesu 4y agoI can't say for sure. Something tells me it would take a paradigm-shift though, the current free-association model is showing it's limitations.