Thought of the day looking at the news: People who sticks unreasonably strongly to a cause cannot be reasonable. And unfortunately there are far more ways of being obnoxiously unreasonable than greatly unreasonable.

BTW what's going on in the key / query space so that distance makes any sense in spite of the curse of dimensionality? Things are highly clustered? Grouped in sub-spaces?

AI challenge: Design a Twitter bot-detector that does not false-positive on Twitter's former CEO.

BTW how can we *not know* who is behind gpt2-chatbot?

Without going to this bizarre type of architecture, I always thought that the variance of the gradient could be exploited to modulate capacity (I tried a bit and failed).

Reminds me a bit of a paper I read (and loved!) three paradigm shifts ago, where they proposed an RL algorithm that detects weights that do not stabilize during training, and replaces them with outputs of newly created units.

Not as a bitter comment, but as a piece of information for those who got papers rejected from ICML: My group got 2 out of 5 and at least one rejected should *really* have been accepted.

You're asking too many questions

Here is the full email. Interesting glimpse at the vulnerability and weaknesses of a mega-corporation that seems (at least to me) invulnerable.

You do not have a shortcut to cut the selected region and automatically paste it in a file fridge.txt in the current directory??? How can you function.

It's crazy how once more, this time in Georgia, unarmed people just want the armed people to leave them alone and not to make their life miserable.

Go Georgia ! #TbilisiProtests #tbilisi

May 1st anywhere on earth.

Thomas G. Dietterich