eachro

A lot of people in SF have nothing to lose.

I remember this one bit from a video on self driving cars back a few years ago where someone basically said that self driving cars would not work well in cities because people will always want to mess with the cars just because they can.

Consider how in the early 2000s when people would mess around with AIM's smarterchild chat bot and gave it outlandish scenarios just to see how it would react. I have to imagine you might see something similar here with these cars but you're right that having a human onboard would probably curtail these interactions.

SF seems like not the best place to pilot this from a risk management perspective (ex: homeless people messing with the cars).

> Each of the big pharma companies were poised to make a boatload of money by being first to produce a vaccine. I'm disappointed, but not surprised.

Considering the massive drop in deaths as various countries, US states got vaccinated, I'd say they haven't been compensated enough.

Just because they were both started by the same guy does not make one a replacement for the other. ESPN ran Grantland into the ground ages before The Ringer was even launched.

Yea grad level absolutely (pretty much anything can fly at the grad level). Undergrad? Maybe we should teach it b/c of the historical importance it has to the field and how the community developed but I really do think most ML classes would be better off without it b/c of the extra background you'd have to use precious time on. Kernel PCA, kernel regression are better for demonstrating the power of kernels.

I suppose the idea of a maximum separating hyperplane is kind of unique to SVMs and if you just teach SVMs through the primal and leave it at that, you don't need to spend all that much time motivating the dual.

To really understand it (primal, dual formulations) you need tools from convex optimization. So it doesn't really feel appropriate to teach it in a standard machine learning class (unless you just toss out the details). In optimization classes, you go through tons of different applications of the methods you learn about: SVM slots in perfectly there. It hits duality, quadratic programming, even gradient descent (Pegasos).

Re deep learning: right, I was bringing up deep learning as a clear example of why you might not want to classify it under optimization. No one considers applications of deep learning to be optimization. However, work on the various optimizers (Adam, adagrad, second order methods, etc) which are all fundamental to doing any deep learning work would be firmly in the field of optimization.

Putting aside discovery, I've always considered SVMs to be in the realm of optimization rather than ML or statistics (though I suppose you could then also put modern deep learning under optimization too).

The code in the statistics library is actually quite a bit more complicated than I had expected. Is there a reason why a function like `mean` isnt just: `sum(data) / len(data)`?

what's the intuition for why that might be desirable? I can sort of see that you might care to consider the relation between a given row and other rows (not disimilar to something like kernel methods) and then you can use something like Deep Sets[1] to featurize the data?

[1] https://arxiv.org/abs/1703.06114

Oh that's interesting. Can you say more? Haven't seen much relating the two topics before.

What is happening on the ground in El Salvador?

Note that the author of this piece was the founder of Coderpad, not some industry entrenched journalist. After watching his interview with Jason Calacanis(https://www.youtube.com/watch?v=5hUT8VZNvm8), where they discuss the content of his piece, I'm more inclined to believe what Woo wrote.

Jason Calacanis's interview with Vincent Woo(the author of that piece) is quite good too: https://www.youtube.com/watch?v=5hUT8VZNvm8

> An average Joe could probably write influential ML papers by gluing RNN/GAN layers to existing models and fiddling with the parameters until they beat current state of the art.

Right, but you have to remember there are legions of grad students doing exactly this so it ends up being quite competitive to churn out papers this way.

If people actually read that Chen/Roose piece they'd see that the conversation being had was much more tempered. I saw the point as about the dangers of being in a large echo chamber.

For ex: on twitter, if a non-private account shares an article that completely misrepresents some statistc, there's a reasonable chance someone will swoop into their replies and push back on it. We see this all the time with both right leaning and left leaning pundits. Does this dynamic go away in large telegram chat rooms? My guess is it still happens in some form (give people some credit, most people are not blind fools). But then again, who really knows unless you're in the private group?

Who is "they" here?

Any list of top rollerball pens cannot be complete without the uniball signo 207(ideally the needlepoint tip): https://uniballco.com/207-needle/

My only complaint about it is that the clip and casing are not great. The Pilot G2 actually has the perfect case but is too inky for my tastes (and stores usually dont carry the 0.38/0.5 widths). The dream pen would have the Pilot G2 case with the uniball signo 207 needle cartridge.

EDIT: It turns out the uniball signo 207 is actually a gel pen, not a rollerball!

What I find fascinating is how people so often have this impulse to tear down things that others hold in high regard. Impulse might even be an understatement - it's almost a need. I wonder why that is.

Ex:

- "vim sucks! why don't people use emacs instead?"

- "katy perry's music sucks!"

Everyone is entitled to their own opinion. Many liked it. Many did not - yourself included. That's okay. Art does not evoke the same response from everyone and we should not expect it to.

This is super minor but I really wish f-string formatting were the default behavior for `print` or that there were a `printf` command just for it. Typing that extra f character before the quotes is just annoying enough

Linux is just 2% of the desktop market. Apple and Microsoft have good enough cloud storage + sync, so Dropbox needs to be even more compellling on macOS/windows to survive.

Their Dropbox Transfer offering is a nice alternate revenue stream for them. I wonder what else they can come up with.

Yup. He could have removed the entire sentence about Parler and the tweet would retain most of its meaning.

Has the author looked at how GNNs are used in say predicting molecular properties[1] or molecule generation[2]? GNNs are the state of the art on those tasks and those problems are certainly much harder than the Cora/Citeseer/etc semisupervised node label prediction tasks.

[1] https://arxiv.org/abs/1704.01212

[2] http://proceedings.mlr.press/v80/jin18a/jin18a.pdf

This is a pretty fun thing to try to do with trading cards. Back in the day when I played Magic/Yugioh, I'd start with a janky common card and see how much I could turn it into. You have to exploit differences in how people value cards: the valuations of cards online mostly follows the way competitive players value them, but there are tons of people who play casually and others who just chase shiny/cool looking cards. It was relatively easy for me to turn a janky card into another common card that was useful in competitive decks (ex: fissure/smashing ground in yugioh), and trade those for janky holos with competitive players, which I could then turn into more useful/valuable holos with the collectible/casual crowd, and so on. It's super fun to do a few times just to know that you can pull it off.

how does one study tactics?

I can imagine a scenario where a politician who wants to pick a fight with Google uses some of the unflattering findings in the published work as supporting material for why Google needs to be regulated/fined/etc: "Google does {bad thing}. Look at this research report from Google researchers! They admit to doing {bad thing}!"

But in this scenario, shielding Google from liability is actually a primary concern given that Timnit is discussing ethics/bias. A paper on say novel transformer architecture, the lottery ticket hypothesis in a new setting, a new RL benchmark suite, etc is not going to expose Google to legal risk the way ethical AI research often can.

I find it funny that the author is concerned about Scale being the kind of company people join for brand rather than "substance". He never mentions the biggest reason people join companies (esp. in SV): to get rich. It is true that undergrads will intern for company's brand and that makes sense b/c just having FAANG on your resume completely changes your job search experience. But most fulltime employees are looking primarily to maximize their pay + RSUs.

Also, as others have mentioned, companies/leaders need to give people something to "give a shit about". For some people that may just be working on "High quality training and validation data for AI applications", but I imagine for most people that's probably not quite enough to make them excited about work. But, who knows? :shrug:

It sounds like a big hassle for you or your accountant.

A large part of the MBA is access to a network of rich/ambitious/well-connected people. The earlier you have your MBA, the more time you have to take advantage of it.