You can add location information to your Tweets, such as your city or precise location, from the web and via third-party applications. You always have the option to delete your Tweet location history. Learn more
Blackbox neural network is not symbolic reasoning. We raised these concerns in openreview of this paper https://openreview.net/forum?id=S1eZYeHFDS … Sadly too much dogma and not enough scientific method in #Deeplearning. You cannot claim to be learning reasoning by overfitting @GaryMarcushttps://twitter.com/ylecun/status/1231346676285280256 …
When test distribution has same complexity as training, there is no extrapolation being tested. A large transformer can learn mapping without learning to reason. We showed extrapolation using compositional models is much better https://arxiv.org/pdf/1911.01545.pdf …
I couldn't have said it betterhttps://twitter.com/basit_ayantunde/status/1231355471606484997?s=19 …
Yeah, this work provides no guarantees on the correctness of the output. Wolfram alpha will either take a very long time or give up, but it won't confidently return the wrong answer like this might
This is actually not a problem in this case, b/c you can differentiate and check equivalence to the original problem in most cases, I believe. The concern is that the success rate of this procedure may precipitate once we deviate from the training data used at the training step.
I share your concerns. But do you think symbolic reasoning can never be achieved by deep learning? Given that DL involves sub-symbolic manipulation, and that symbolic manipulation is a special case? Couldn't it be that the problem lies with training in DL and not DL itself?
would urge you to read both https://arxiv.org/abs/1912.05752 and my arxiv https://arxiv.org/abs/2002.06177
Deep learning and symbolic reasoning both fall very short in replicating the reasoning needed by safety-critical systems w/ embedded AI. What we really need is for *famous* AI academics and practitioners to stop posturing and solve a very real problem http://hal.pratt.duke.edu/sites/hal.pratt.duke.edu/files/u39/2020-min.pdf …
my suggestions here: https://arxiv.org/abs/2002.06177
and see also https://arxiv.org/abs/1912.05752
Meanwhile, in blackbox-is-evil-and-no-results-can-be-trusted-land... https://www.schneier.com/blog/archives/2013/02/guessing_smart.html …
Twitter may be over capacity or experiencing a momentary hiccup. Try again or visit Twitter Status for more information.
