10 years at Google this week.. how time flies.
There’s a lot of impressive stuff about this company but the top notch people are what consistently stand out to me.
Thank you to everyone who made me who I am over the years!
Once you realize that the essence of good software is well defined API boundaries you start to realize that’s also the essence of all good human relationships too.
Regardless of your opinion of value / efficacy / issues, this has been a WILD week for machine learning.
From audio synthesis in your DAW (mawf, RAVE) to PaLM to robotic control with natural language.
There are still major walls but the process is still exciting!
A reminder that we build huge models and then distill / shrink them later, but scaling first seems to be a requirement for this process that seems to work well (simpler to see what’s possible).
See inception => mobilenet, encoder decoder translation networks, facemesh, Bert, etc
I’m much more interested in the human side of software engineering than the technical.
This is an art: constructive criticism, team building, how to choose the right people to work with, how to stay happy long term.
All “unsolved” problems, in some sense.
surprise trip to the south bay for a couple days, first time since covid!
DM me if you meet up, I’m ripping around in a white ‘14 boxter with an aftermarket stereo :)
It’s important to always ask why you’re following industry norms — they’re meant to be guidelines, not dogma.
My philosophy is do whatever makes sense for your project — be pragmatic. You can break any rule if it makes sense for your project.
I’m going to say this one more time …
You cannot have Continuous Integration with feature branches.
You can have automated builds. Which is a good thing. But not CI! Alright?
You might have CI with short-lived branches of a couple of hours.
"[T]he Office will not register works produced by a machine or mere mechanical process that operates randomly or automatically without any creative input or intervention from a human author."
Machine creates z-vector, human adds 1e-10.
Is this okay?
"It's a real shame that this is the world we find ourselves in now. One so sharply divided by politics and ideology that we can't even share the love of Ruby on Rails together at a conference without a need to settle scores." https://world.hey.com/dhh/no-railsconf-faa7935e…
Pocket cameras & internet seem like the most important technical inventions of the last few decades.
The war in Iraq is probably the last war that didn’t have this type of worldwide visibility.
I have a really hard time understanding how sanctions are going to protect a modern city where families are hiding in the subway and now instructed to make guerrilla molotov cocktails, one we vowed security assurances in 1994.
This is all we can come up with?
Work on the “dumb” problems. Things that make people go “well that’s obvious isn’t it?” but hasn’t been done properly.
Usually those things are timeless :)
We computer vision researchers rarely look at the individual data points inside our datasets.
Mainly because we are too lazy and/or do not have the right tools.
This needs to change. And now we have a great tool from
I used a neural network to source separate 50 cent vocals from instrumental, and used neural style transfer on the preview art
oh and hello my new musical pseudonym bitflippa
50 Cent - Disco Inferno (bitflippa remix)
https://soundcloud.app.goo.gl/ecbFpHUFiJHMHD4Z8…
Design docs are best written retroactively.
Until a working prototype is built a design doc has little value since it’s a conversation about hypotheticals.