> but I think most ML people now think of neural-network architectures as being, essentially, choices of tradeoffs that facilitate learning in one context or another when data and compute are in short supply, but not as being fundamental to learning.
I feel like you are downplaying the importance of architecture. I never read the bitter lesson, but I have always heard more as a comment on embedding knowledge into models instead of making them to just scale with data. We know algorithmic improvement is very important to scale NNs (see https://www.semanticscholar.org/paper/Measuring-the-Algorith...). You can't scale an architecture that has catastrophic forgetting embedded in it. It is not really a matter of tradeoffs, some are really worse in all aspects. What I agree is just that architectures that scale better with data and compute do better. And sure, you can say that smaller architectures are better for smaller problems, but then the framing with the bitter lesson makes less sense.
Not what happened though. Orkut actually launched a month before facebook and it was a 20% project, it wasn't a panick move or a big project for them. And it was great, imo google did badly on not maintaining it.
You can make this point about google+ though, totally panick move.
I watched two episodes of Komi Can't Communicate on Netflix and felt that. I thought it was good though, maybe it would have helped my socially anxious younger self.
A counterpoint is that at least on reddit the downvote button is rarely well used. It's always a "I disagree" button, so both bad and controversial things are pushed down. It's not the case here in general I think, but I actually think this is a good argument in favor of removing dislikes. It filters some bad things but also just controversial things for one reason or another.
It's happening on Brazil I think. It's very hard to hire for brazilian companies since people can earn an extraordinary salary by working remote. Our timezone is also pretty close to the US east coast, and still not that far from SV.
That also looks like a cool project! I've made a bot to play a game called Gems of War. I've used some GUI windows thing to program the AI but it was terribly limiting and slow. I was going to start again using python and opencv from scratch but I was kinda sad I needed to code a lot of plumbing code such as finding the screen, and then finding sprites and things like that. Your project seems to do that and would save me a lot of hassle. I may try again someday.
It's not bullshit. But at the same time, I wonder why so many people do these majors. They mostly end up working on a unrelated field. If they go to university just because they are passionate about learning the subject and are willing to pay for it, that's great! But I feel it's just that they are pressured to go to college, they are pressured to "follow their passions" even when it will not work out, and in fact some menial jobs require a degree nowadays.
We don't really need (economically speaking) lots of philosophy majors. And maybe that's okay if people are doing philosophy majors for the knowledge, but I don't think that's the case. And if they want a good job, I think we should discourage these majors.
That doesn't mean we don't need philosophy. It's really important. But it doesn't make much difference if the people who have a degree on that work on unrelated things.
> Larger lunch tables were "driving more than a 10% difference in performances". A fact that would probably have gone undetected without such data analysis.
Isn't this poor data analysis, though? Correlation doesn't imply causation and all that. I think it's more plausible that people who sit at 12-person lunch table have more friends on the company, and that drives performance up, than the lunch table affecting performance. I mean, it could be any reason, but I highly doubt the lunch table is the cause of the higher or lower performance.
Wow, so cool, but so confusing. I can't figure how that would work. But I also can't figure how people can think of a cube by its properties instead of seeing it. This gives me lots of food for thought. I will probably try shutting up the inner voice and projector later to try to understand it better.
When people talk about an "inner voice". Like when you did something bad and you berate yourself mentally for it. Don't you hear a voice? Or when thinking about an imaginary argument with someone? Don't you do that?
I don't know much about control, but apparently Emanuel Todorov knows about the basics of Boston Dynamics control in this video https://www.youtube.com/watch?v=7enj1FGoYwg around 14 minutes. The paper is this one https://repository.upenn.edu/ese_papers/686/. I don't understand most of it, but it may help if you have some knowledge. But from the few things I understood, not much is programmed.
Also, I'm watching some lectures for a MIT underactuated robotics class (https://www.youtube.com/watch?v=VeEqtTgDXFc). The professor says in one of the videos that the people who end up creating Boston Dynamics also took that course. From what I watched until now, nothing is explicitly programmed (by some vague definition of "explicitly programmed"), you just find some inputs that makes the dynamics of the model behave the way you want.
Almost certainly the birdlike movement is the outcome of the robot dynamics, and not some hardcoded rules. But things like avoiding hitting obstacles or estimating the distance, I think that needs to be explicitly programmed (and then used as input to the model such as "move the end effector to that position").
I tried it on Firefox and it works fine! I just needed to change the value of <secret_key> to a key I created on their website (or else it fails silently, it would be cool to have some console message I think).
The first article is really anecdotal and there are a lot of other explanations, which the commenters talk about.
Also, the illegality of LSD makes it so regularly people use something else thinking it's LSD. Part of the bad problems (such as HPPD) related to LSD usage could be caused by something else. Could also explain why it seems mushrooms are safer.
But it could also be that hallucinogen usage make people weird, yeah. As you said, we don't really know yet. I'm curious and really hopeful.
>I think the psychedelics nurture the ability to recognize that feelings and narratives are not representative of reality
>The ability to detach from self, to recognize a spectrum of truth, of perception
It sounds a lot like meditation and buddhist practice. I mean, I already hear that everywhere and there are some clinical trials with meditation too. But your narrative fits nicely. That's cool.
Having Alan Turing as a role model helped me immensely with internalized homophobia.
When I was younger I honestly thought there weren't many intelligent gay man, and that it was somehow a proof that I was never going to be really good. That was until I discovered Alan Turing was gay, and actively researched about intelligent gay people.
Role models that are like you are so important. One day a friend asked why I got excited when I discovered some guy I admired was gay, and then I realized that he doesn't understand because he has lots of people like him in any area. I know it sounds kind of tribalist in a way, but when society puts you down because you are in a certain group, it's really hard to avoid thinking you are "limited" because of who you are. After seeing lots of people like me, it's less important to me to have these role models since now I know I'm no less of a person because of who I am.
Point is, I don't think most people understand the impact of having role models that are like you, and that's why they don't "get it" when people talk about it.
I'm not in Toptal, but a friend is and he asks for around $5k a month. Here in Brazil it's insanely good. I don't personally know anyone else who receives more than him here, actually. In software dev or anything else. I don't go around asking how much people make, though.
I feel like you are downplaying the importance of architecture. I never read the bitter lesson, but I have always heard more as a comment on embedding knowledge into models instead of making them to just scale with data. We know algorithmic improvement is very important to scale NNs (see https://www.semanticscholar.org/paper/Measuring-the-Algorith...). You can't scale an architecture that has catastrophic forgetting embedded in it. It is not really a matter of tradeoffs, some are really worse in all aspects. What I agree is just that architectures that scale better with data and compute do better. And sure, you can say that smaller architectures are better for smaller problems, but then the framing with the bitter lesson makes less sense.