You win by acknowledging that AI/Machine Learning research has existed long before this bubble existed, and is continuing to happen outside of that bubble. Most of what we call AI nowadays is based on neural networks (that’s what Geoffrey Hinton and others got a recent nobel prize for), but that’s not the only way to go about the problem, and for years now, there have been researchers pointing out problems like hallucinations and diminishing returns from increasing the amount of data you feed to a model.
An example of one such researcher is Song Chun-Zhu, who has recently moved back to China because he was finding it increasingly difficult to do research he wanted (i.e. outside of the current AI bubble) within the US. That linked article is a bit of a puff piece, in that it is a tad too mythologising of him, but I think he’s a good example of what productive AI research looks like — especially because he used to work on the “big data” kind of AI, before realising its inherent limits and readjusting his approach accordingly.
He’s one of the names that’s on my watch list because even for people who aren’t directly building on his research, he comes up a lot in research that is also burnt out on neural nets
How do you “win” a tech / investment bubble? Is it a race to see whose economy implodes first?
You win by acknowledging that AI/Machine Learning research has existed long before this bubble existed, and is continuing to happen outside of that bubble. Most of what we call AI nowadays is based on neural networks (that’s what Geoffrey Hinton and others got a recent nobel prize for), but that’s not the only way to go about the problem, and for years now, there have been researchers pointing out problems like hallucinations and diminishing returns from increasing the amount of data you feed to a model.
An example of one such researcher is Song Chun-Zhu, who has recently moved back to China because he was finding it increasingly difficult to do research he wanted (i.e. outside of the current AI bubble) within the US. That linked article is a bit of a puff piece, in that it is a tad too mythologising of him, but I think he’s a good example of what productive AI research looks like — especially because he used to work on the “big data” kind of AI, before realising its inherent limits and readjusting his approach accordingly.
He’s one of the names that’s on my watch list because even for people who aren’t directly building on his research, he comes up a lot in research that is also burnt out on neural nets