brilee

9 ms·

61.
Bureaucratic Leverage(moderndescartes.com)
1 points by brilee 3y ago|0 comments
62.
by brilee 3y ago
Most likely it has badly conditioned embedding vectors for those particular tokens, leading the network to edge into numerically unstable territory; once you get some sort of underflow or NaN, they tend to propagate and invalidate the entir
63.
by brilee 3y ago
I'm sure you're aware that ferrocyanide is not the same thing as cyanide - the cyanide is bound so tightly to the iron center that it is nontoxic.
64.
by brilee 3y ago
https://www.moderndescartes.com/essays/factobattery/ I looked at a variety of these ideas and most of them are bad. Desalination is another one I looked at quickly (did not add it to the blog post), but I recall i
65.
Bureaucratic Leverage(moderndescartes.com)
1 points by brilee 3y ago|0 comments
66.
by brilee 3y ago
geohot built one of these in high school https://www.forbes.com/forbes/2007/0723/060.html?sh=3c183bcb...
67.
How to Be a Good Codemate(moderndescartes.com)
2 points by brilee 3y ago|1 comments
68.
How to Be a Good Codemate(moderndescartes.com)
1 points by brilee 3y ago|0 comments
69.
by brilee 3y ago
In this long rant, there is only one concrete thing named that is supposedly bad: > I saw someone that was working on a few Git branches, twenty years my senior, and they decided to clone the entire repository ten times, then checkout a
70.
by brilee 3y ago
see https://www.rdkit.org/docs/GettingStartedInPython.html#morga...
71.
by brilee 3y ago
The advice says keep 25%, but it also says to rebalance every year. In a year when stocks are down, cash becomes a larger fraction of your portfolio and you would use it to buy stocks precisely when they are at their lows! On the flip side,
72.
Readability: Google's Temple to Engineering Excellence(moderndescartes.com)
2 points by brilee 3y ago|0 comments
73.
by brilee 3y ago
The heat generated by drag at such high speeds means that titanium was, at the time, the only suitable material due to is high melting temperature.
74.
by brilee 3y ago
I think you misunderstood the comment. I agree there has to be a buffer somewhere. You're arguing that we can get a free buffer by piping the water behind a reservoir, but the cost is that the very capital-expensive desalination plant
75.
by brilee 3y ago
https://www.moderndescartes.com/essays/factobattery/ Using the above methodology, I ran numbers for a desalination factobattery and it's really bad, like two orders of magnitude more expensive than just buyin
76.
by brilee 3y ago
Almost. You would have to maintain your underwater base at 1atm and you would indeed get free freshwater. The catch is that the internal volume of your underwater base would decrease and you would eventually have to eject wastewater by push
77.
Readability: Google's Temple to Engineering Excellence(moderndescartes.com)
4 points by brilee 3y ago|0 comments
78.
by brilee 3y ago
www.moderndescartes.com/essays Pretty much all over the board, with infrequent but polished essays
79.
Vectorizing Graph Neural Networks (2020)(moderndescartes.com)
74 points by brilee 3y ago|15 comments
80.
by brilee 3y ago
The U.S. quite appropriately cut its defense spending after the USSR collapsed: https://www.macrotrends.net/countries/USA/united-states/mili... which in turn caused a cascade of mergers in defense contractors
81.
by brilee 3y ago
I've never seen DeepMind do a press release without something seriously impressive to announce. This sort of insignificant chitchat with the press seems like he got pressured by higher ups to talk about LLMs for stock price reasons.
82.
by brilee 3y ago
I've been quite careful not to divulge anything confidential, and anything that is remotely close to sensitive has publicly accessible citations. My opinions about Google are tantamount to discussions about workplace conditions, and it
83.
by brilee 3y ago
You're right that I don't know enough as an IC to comment on the thinking, actions, and promises at the exec level that kept Brain funded. But even if I did, to the GP's point, I would not talk about that part at all. I'
84.
by brilee 3y ago
sorry, I had a mind fart. I was thinking of seq2seq https://research.google/pubs/pub43155/ Pushing the fix now...
85.
Why did Google Brain exist?(moderndescartes.com)
502 points by brilee 3y ago|315 comments
86.
by brilee 3y ago
This is untrue. Ice is about 91% the density of water. If you have water at 0 height, and you put in water equivalent to +1 unit of height. If it's in liquid form, obviously height goes up by +1. If it's in ice form, there is +1.0
87.
by brilee 4y ago
Today, compute speed is roughly 5-6 orders of magnitude faster (single threaded) than it was 30 years ago. Today, RAM is roughly 5-6 orders of magnitude faster than HDD was 30 years ago. All the old-school HDD-optimization techniques from t
88.
by brilee 4y ago
https://en.wikipedia.org/wiki/Magnetohydrodynamics Sure we do. It is different and no simpler than modeling weather. Weather at least has the advantage of a 3D system decomposing into a 2D sphere X 1D height dimension
89.
by brilee 4y ago
"There are only two ways to make money in business: bundling and unbundling"
90.
by brilee 4y ago
Oh wow, that's amazing. Thanks for the clarification. Glad you got involved with the paper!

More ›