2 ms·

Since I've been seeing this wild fantasy being bandied about for a while now, I have to point out that, if you had "AI" that could train "AI", you wouldn't need

Since I've been seeing this wild fantasy being bandied about for a while now, I have to point out that, if you had "AI" that could train "AI", you wouldn't need to train any more "AI". Because at that point, there would be nothing to gain. Suppose you have a text classifier that can produce text classifications just as good as those of human annotators, so that you could use it to train other text classifiers. At that point, why would you need to train other text classifiers? What changes to predictive accuracy would you expect to achieve? Or suppose you had a language model that could produce language just as well as humans (not just in terms of grammaticality, but also in terms of making sense, I guess?). At that point, why would you need to use the text generated by that language model to train another language model? What kind of changes in the quality of the new model's generated text would you expect to see? Same goes for image classifiers, and any other classifiers, or generative models you may care to consider. Note that both language models and image classifiers have been "beating" human performance in benchmarks for a while now, and still they are not used to train other classifiers. And the reason for that is that it doesn't make sense: the ground truth is always the decisions made by humans. There is nothing in machine learning theory or practice that says that a classifier can perform better than whatever process originally labelled its data. If it looks like it does, it's because there's something wrong with the methodology (i.e. overfitting, bad benchmarks, some other kind of nonsense).

sebzim4500 4y ago
>if you had "AI" that could train "AI", you wouldn't need to train any more "AI". Because at that point, there would be nothing to gain. Not necessarily, the AI used to classify might be much more expensive to run than the new AI you are training. This is what Tesla does, for example. They have a massive model which they use as a 'ground truth' to train the models which can actually run in the car.
- YeGoblynQueenne 4y ago
  Well, OK, there are gains in efficiency, but the context here is, as far as I understand it, letting an "AI" train a better "AI". Perhaps I misunderstand the OP and like I say in my comment above, I've seen comments to that extent pop up here and there om HN in the last few days, so maybe I'm jumping to conclusions about what the OP meant. I should have asked for clarifications (although usually when I do there's no response; not from the OP, from most users).
nico 4y ago
> Note that both language models and image classifiers have been "beating" human performance in benchmarks for a while now, and still they are not used to train other classifiers. Mmhh have you seen Alpaca? It’s the Llama model (Meta’s), trained (fine-tuned) using GPT. Yesterday there was a paper showing ChatGPT is better than mechanical turks. > the ground truth is always the decisions made by humans But why does AI need humans’ interpretations of reality anyway? AI can just get data from the real world instead and use its own intelligence to make decisions. It doesn’t need to learn from humans anymore. We just want it to. Things are going faster than we can keep up with, and accelerating. Barring a catastrophe, AI is not going to stop, and soon it’s not going to need humans to keep it going anymore.
- YeGoblynQueenne 4y ago
  I replied the way I replied because I assumed you meant "AI" training better "AI". Please help me understand if that's what you really meant, because if that's what you meant then, yes, that is a complete fantasy.

sebzim4500 4y ago

>if you had "AI" that could train "AI", you wouldn't need to train any more "AI". Because at that point, there would be nothing to gain. Not necessarily, the AI used to classify might be much more expensive to run than the new AI you are training. This is what Tesla does, for example. They have a massive model which they use as a 'ground truth' to train the models which can actually run in the car.

YeGoblynQueenne 4y ago
Well, OK, there are gains in efficiency, but the context here is, as far as I understand it, letting an "AI" train a better "AI". Perhaps I misunderstand the OP and like I say in my comment above, I've seen comments to that extent pop up here and there om HN in the last few days, so maybe I'm jumping to conclusions about what the OP meant. I should have asked for clarifications (although usually when I do there's no response; not from the OP, from most users).

nico 4y ago

> Note that both language models and image classifiers have been "beating" human performance in benchmarks for a while now, and still they are not used to train other classifiers. Mmhh have you seen Alpaca? It’s the Llama model (Meta’s), trained (fine-tuned) using GPT. Yesterday there was a paper showing ChatGPT is better than mechanical turks. > the ground truth is always the decisions made by humans But why does AI need humans’ interpretations of reality anyway? AI can just get data from the real world instead and use its own intelligence to make decisions. It doesn’t need to learn from humans anymore. We just want it to. Things are going faster than we can keep up with, and accelerating. Barring a catastrophe, AI is not going to stop, and soon it’s not going to need humans to keep it going anymore.

YeGoblynQueenne 4y ago
I replied the way I replied because I assumed you meant "AI" training better "AI". Please help me understand if that's what you really meant, because if that's what you meant then, yes, that is a complete fantasy.