ericjang

Let someone else waste the valuable resource :)

Auction the team as a resource to the highest bidder. I'm not positioned to deploy this resource effectively, so I'm likely to benefit most by trading it to a buyer who can't convert money into talent fast enough.

The team is much more valuable to someone who can afford to pay long-term salaries and needs this talent resource quickly on an up-front basis (Tesla, Uber, a startup trying to crush their competitor, etc.)

can you elaborate on what you disagree about?

I think the last remark about Xi Jinping's wife was uncalled for (I'm Taiwanese American).

more progress coming soon. You aint see nothing yet ;)

The authors of the blog post make pretty liberal use of fantasy artwork. The art doesn't add any educational value for me. If I stretch my imagination I can see how the last image gives a connotation of "an impressive, highly skilled, well refined entity", but to be it's... cringey for a company to be doing this. Like, I see my high school self in this aesthetic.

Also, the authors would do well to credit the original sources / authors.

> The world is full of hidden complexity, so be prepared for an education at any moment, I'd say.

Wonderful words! Added to my quotes list :)

There is at least one Google researcher on that list (and various academic professors on that list are employed by Google as consultants).

maybe she/he

Pixar does not have an open floor plan. It's 2-3 folks to an office, and the animators are well known for decorating their offices in fairly quirky ways during the "off-season" when they have finished their portion of the work for a movie (https://www.google.com/search?q=pixar+animator+offices&sourc...)

Pardon the naive question, but is a high-rise corporate office like this one (and firms in NYC) preferable to a sprawling "groundscraper" campus (Google, FB)? Which layout costs less? How does this affect management and organizational structure?

Tacotron 2 text-to-audio samples:

https://google.github.io/tacotron/publications/tacotron2/ind...

Blog post: https://research.googleblog.com/2017/12/tacotron-2-generatin... Paper: https://arxiv.org/abs/1712.05884

"Take back control of your content and unlock your sources of inspiration for just 5€/ Month."

Sounds like a tagline for a "Silicon Valley" episode.

+1, definitely well worth reading. Paragraphs 85-89 seem like they come out of a prize-winning thriller, but every single paragraph in the indictment is also gripping.

Here's a recent survey / observational science paper by some prominent "neuroevolution" / A-Life researchers. https://arxiv.org/abs/1803.03453. I found this refreshing because it's rare that science papers talk about the debugging and experimental process and debugging journeys underlying this research.

The author has a pretty good sense of humor. On the testimonials page (http://lcui.org):

"LCUI solves many complex issues, it reduces the application UI development and maintenance costs, I only need to write a code for the UI will be able to make my application in a number of platforms to show a consistent UI effect." -- Liu Chao, Creator of LC's Finder

I used gRPC for numerous hobby projects during my undergrad to glue together binaries running in different languages (e.g. a simulation server running in C++ and a scripting client in Python). By passing around a shared data structure (Protobufs), one does not need to waste time writing serialization/de-serialization adapters. It is also useful for gluing together microservices.

FB's Thrift also solves the same problem, and is an alternative to gRPC.

Out of curiosity - how do people who employ these marketing tactics reconcile this? do they drink their own Kool-Aid or do they willfully turn a blind eye to the grey/blackhat clients because they don't want to think about it?

It was started by demoscene people.

Thank you for this comment! Gives me a lot to think about. Bookmarked.

Nice. I also recommend checking out Nanogui (C++) by Wenzel Jakob: https://github.com/wjakob/nanogui

Thanks for the good science!

In vanilla policy gradient, one plays the game to the end and then bumps the probability of all actions taken by the agent up (if AlphaGo won) or down (if it lost). This is very slow because there ~150 moves in an expert game, and we do not know which moves caused decisive victory or loss - i.e. the problem of "long term credit assignment". Also, it is actually preferable to compute the policy gradient with respect to the advantage of every action, so we encourage actions that were better than average and punish actions that were worse than average - otherwise the policy gradient estimator has high variance.

I think about MCTS in the following way: suppose you have a perfect "simulator" for some reinforcement learning task you are trying to accomplish (i.e. real-world robot grasping for a cup). Then instead of trying to grasp the cup over and over again, you can just try/"plan" in simulation until you arrive at a motion plan that picks up the cup.

MCTS is exactly a "planning" module, and it works so well in Go because the simulator fidelity is perfect. AlphaGo can't model adversary behavior perfectly, but MCTS and the policy network complement each other because the policy reduces the search space of MCTS. As long as the best adversary is not far away from the space that MCTS + policy is able to consider, AlphaGo can match or beat the adversary. Then, we train the value network to amortize the computation of the MTCS operator (via Bellman equality). Finally, self-play is an elegant solution for keeping adversary + policy close to each other.

For more rigorous mathematical intuition, Ferenc Huszar has a nice blog post on MCTS as a "policy improvement operator": http://www.inference.vc/alphago-zero-policy-improvement-and-...

Money is no object to HFT firms when it comes to reducing latency. A recent technique involves laying optic fiber cables with hollow interiors, so that the internal refraction of light spends more time in a vacuum and less time slowed down by the pesky glass medium.

I wish we knew how to imbue robots with chemical actuation mechanisms that actually worked (as opposed to some proof of concept demo of a simple electrolysis reaction in a bulky 3D printed silicone balloon).

If real muscle cells / limbs serve as a sort of reference, they would be vastly more space-efficient, compliant, and have higher power-weight ratio than any electromagnetically actuated servo we have today.

It makes me wonder if the "modern" androids in Westworld used electricity at all.

I see, so you are proposing a way to infer capillary expansion/contraction as a way to measure blood flow.

Does de-oxygenated blood have a measurably different refractive index w.r.t. radio waves than oxygenated blood?

The analysis in this article was based on data from Thomas Piketty and collaborators, who have also compiled an extremely detailed and comprehensive analysis of the subject of capital & wealth & inequality. I highly recommend Piketty's book, "Capital in the 21st Century" (see Goodreads comments for abridged summaries and reviews by readers). Not only is it totally convincing (and depressing), but it also provides a very clear understanding of the course of modern history since the 19th century. It's amazing to see how clear economic policy (whether Marxist or Capitalist or anything in between) becomes when viewed through the lens of society grappling with the problem of capital/income/inequality.

https://www.goodreads.com/book/show/18736925-capital-in-the-...

Why are you so bullish in crypto? Would love to hear your thoughts here (or send me a DM if you prefer).

This. I wrote a crypto arb bot in 2014 and decided to stop working on it / open-source it for the same reason after a couple exchanges that I had cryptos in mysteriously vanished.