François CholletVerified account

maybe because back-propagation is not the most efficient algorithm? Is the world waiting for something better? #deeplearning #ArtificialIntelligence

Probably from a scientific perspective but from an engineering perspective we just got started. Eg voice rec still requires very directed speech to devices and cannot follow in natural settings. Same for vision.

is there a source / some facts that support the "entered diminishing returns territory" argument?

Now we just have to understand how to make these things work efficiently and robustly and develop standardized processes to deal with their development and maintenance. Not that simple...

Means ? Can you explain this a bit ? Are you saying the ROI is not there anymore ?

Perhaps the next wave is simply a further miniaturization and integrating application into all aspects of society.

These areall perception related tasks. Not much happening in common sense reasoning for example

Deep learning systems have no idea what they're seeing. We have to tell them with a label. But they still have no idea. Deep learning is irrelevant to AGI research. In fact, it's a hindrance because it's syphoning funds away from other approaches that hold much better promise.