Great. OpenAI Microscope. Essentially a repost of 2017 post from here https://distill.pub/2017/feature-visualization/ … Nice visuals, lot of fluff, not much science. What do we learn from looking at these images (again, 3 years later)? Did we ever learn anything from looking at this stuff?https://twitter.com/OpenAI/status/1250088291304534016 …
you recognize these are both works (at least in part) by Chris Olah right? this is a pretty big step up from the early work
-
-
Yes I do. I'm not sure about how big of a step it is though. All I see are pretty much same pictures and same pseudo-profound observations such as "yay we found an edge feature in a deep net with 50k neurons".
-
Instead of trying to explain DNNs,
@OpenAI should be trying to devise a new AI model. The fatal flaw with DL is this: The brain can instantly perceive any object without recognizing it. By contrast, a DNN is blind to an object unless it has been trained to recognize it. - Show replies
New conversation -
Loading seems to be taking a while.
Twitter may be over capacity or experiencing a momentary hiccup. Try again or visit Twitter Status for more information.