rryan

> No one at Google is responsible for these half baked and largely irrelevant widgets or wants to stake their career fixing them.

You're just ... wrong about this. There's an entire team of dozens of people (maybe hundreds now) focusing on this specific web answer feature. I personally worked on the team (not this feature, though).

I don't understand why people say things they know nothing about.

AIUI, AssemblyScript is a TypeScript-like language that is designed to compile to wasm.

I love that Firefox exists and Quantum is an amazing step forward, but Firefox still regularly runs away with gigabytes of RAM and hung worker processes. I have no problem with long-lived Chrome sessions but I need to restart Firefox ~daily. It's not bad memories of 10-years ago.

Uh, 2040? Seattle will be underwater by then.

The grandma stereotype is discriminatory towards both age and gender.

Cloud TPU pods are seriously amazing. I'm a researcher at Google working on speech synthesis, and they allow me to flexibly trade off resource usage vs. time to results with nearly linear scaling due to the insanely fast interconnect. TPUs are already fast (non-pods, i.e. 8 TPU cores are 10x faster for my task than 8 V100s) but having pods open up new possibilities I couldn't build easily with GPUs. As a silly example, I can easily train on a batch size of 16k (typical batch size on one GPU is 32) if I want to by using one of the larger pod sizes, and it's about as fast as my usual batch size as long as the batch size per TPU core stays constant. Getting TPU pod quota was easily the single biggest productivity speedup my team has ever had.

Travis has weird priorities. The Ubuntu environment has been stuck on Precise/Trusty for years now. And the only place you can upload artifacts is Amazon S3. I really hate to be that person who sees a new offering and is like "but when will you fix your busted junk?!" but.. when will you fix your busted junk?!

I would kind of expect that "in 2018" we'd be using QML and not QtWidgets?

Loud swearing doesn't show up on metrics. This drives me insane about Amazon. My behavior won't change, even though my perception of the brand is going down the tubes.

Netflix is also horrible about autoplaying trailers with sound. Maybe I went to netflix.com to check whether something was in the catalog while my spouse sleeps next to me. You do not have the right to pump sounds through my speakers until I hit play.

dat: https://datproject.org/

We're coming from Freenode as our only real-time communication so the difference is night and day. Slack is a no-go for many due to not being FLOSS and I'm concerned about vendor lock-in if they were to stop being so generous or run out money and shut down.

Slack's threading model is much worse than Zephyr/Zulip's IMO. Especially on mobile. The streams/topics flow in the "all conversations" view is an incredibly intuitive way to keep track of everything that is going on.

They offer free hosting for open source projects. It has been really great for Mixxx (floss DJ software), allowing our ~10 person distributed dev team to work together much more effectively.

Eh probably not, Zulip was acquired by Dropbox in 2014, long before Slack was popular.

> This technology has been around for a year but we only got a few samples. I'm very excited. I use TTS to read back all the text I consume on PC.

Nope! These two papers are fresh work on prosody modeling. You can see the evolution of work this team has been publishing about here:

https://google.github.io/tacotron/

> This web demo allows you to enter your own text:

That web demo is unrelated to this. It's about a Google Cloud TTS API, which only includes WaveNet, not Tacotron.

https://cloudplatform.googleblog.com/2018/03/introducing-Clo...

It's obvious the topic of this post is strategic -- it will drive traffic which helps him sell his content.

Do StarCraft AIs (the ones that are included with the game) cheat? (e.g. can they see past the fog of war?)

Yup! The code is here: https://github.com/tensorflow/tensorflow/tree/master/tensorf...

Anyone interested in machine learning for audio (or any other signals) should keep a close eye on tf.contrib.signal. I just checked in some code to do efficient framing, windowing and overlap-add. In a day or so I'll be checking in an STFT and inverse STFT with GPU and gradient support, so you can make an STFT part of learning (not just an input pre-processing step). Please file bugs / feature requests and feel free to CC me (also rryan on github).

That's really odd that that Keras API's interval is measured in epochs (which is a different wallclock interval for every different model/dataset/hardware configuration). It's much more common to checkpoint based on a time interval.

/That's/ the lede you're taking away from this story? Some asshole's fall from grace?

Anyone who loves the Lisp concept of "code is data" will love TensorFlow.

Instead of coding imperatively, you write code to build a computation graph. The graph is a data structure that fully describes the computation you want to perform (e.g. training or inference of a machine learning model).

* That graph can be executed immediately, or stored for later.

* Since it's a serializable data structure, you can version it quite easily.

* You can deploy it to production without production having to depend on ANY of the code that built the graph, only the runtime necessary to execute it.

* You can run a compiler on it (such as XLA or TensorFlow's built in graph rewriter) to produce a more efficient version of the graph.

* In some circumstances, you can even compile the runtime away, producing a single .h/.o that you can link directly into e.g. a mobile app.

It's a beautiful and highly useful abstraction that allows TensorFlow to have both a great development and production story in one framework. Most frameworks only have a good story for either development or production.

If you are a machine learning researcher who doesn't need or care about deploying your work (i.e. mostly publishing papers), you may not want the overhead of having to deal with building a graph, and may prefer something that computes imperatively like PyTorch. If you are building products / services that use ML and developing/training your own models (as opposed to taking pre-trained models and using them), there is really no credible competitor to TensorFlow.

Disclaimer: I work at Google. I spend all day writing TensorFlow models. I'm not on the TensorFlow team nor do I speak for them or Google.

> I feel the right to complain anytime

Most people on the Internet do.

Allowing LD_LIBRARY_PATH in production is a potential security risk.

This is one of zsh's many killer features. Years of accumulated shell history with backsearch-by-regex saves me so much time.

No, just put the following in ~/.zshrc:

  # share history among terminals
  setopt share_history

  # append to the history instead of overwriting
  setopt append_history

  # append to history incrementally instead of when the shell exits
  setopt inc_append_history

Look ma, no more lost history!

Emacs user with hundreds of Chrome tabs open? This extension will change your life. It's iswitchb, for tabs.

https://chrome.google.com/webstore/detail/powerswitch/lljbpn...

I just switched from C-z to C-q and couldn't be happier. C-z was fine, but I realized it was a really unergonomic stretch. With caps lock as ctrl, C-q is only a slight movement of my ring finger.

GANs are a general tool -- they just happen to get a lot of attention for generating images of stuff. Here's an example for generating sequences [1]. The example is language oriented, but ultimately GANs are interesting because you can use them to build a generator for an arbitrary data distribution. This can have many applications in engineering (to take a random example -- generating plausible looking chemical structures under a certain set of constraints). As with any ML application, you need to quantify your tolerance for "inaccuracy" (in a generative setting, how well the generated distribution matches the true data distribution). This is simply an engineering trade-off and will vary based on the application.

[1] https://arxiv.org/abs/1609.05473

The person who replaced Amit as head of search (John Giannandrea) is the founder / CEO of Metaweb and creator of Freebase -- aka what became Google's Knowledge Graph. Google using more ML in search doesn't mean a KG isn't also useful for query understanding.

And KGs aren't mutually exclusive with ML -- e.g. https://arxiv.org/abs/1612.04844

> There's a number of things that have happened in the organic search area of Google that seem to suggest a declining interest in quality organic results.

I think you may be falling for the SV cult of personality fallacy. There are still literally thousands of individuals you've never heard of working on organic search quality at Google who care very much about quality organic results. :)