codesushi42

And now we have been blessed with VR chat more than 20 years later.

Uuh, AI research has nothing to do with coding all nighters. This is a common misconception among software engineers. It is more a science, and less an engineering problem. It is more about running experiments than it is writing fancy algorithms.

You are bound by the amount of data and computational resources you have at your disposal. Neither are tied to man hours. You can stay up all night for days waiting for your model to train, and it will do you no good.

This would be great if only he was not at Facebook.

Is the goal of the AI model to predict crime rates in a hypothetical world where everyone has equal rates of lead exposure?

What a terrible example. There are many other features being omitted that would predict crime rates. If anything, this is an example of enforcing your own bias on the model by not including all relevant features.

Why is it that a lot of startups seem to be betting the farm, so to speak, on "machine learning" as their core value

Because it is the current overhyped thing, and there are enough dumb investors to fund it.

My brain.

You will overfit an NN trained on only 1000 examples.

Also a simple train/test split will tell you that. But the author failed to take any time to learn the basics before spewing out this drivel.

McDonal's for someone visiting Europe makes no sense at all.

McDonalds is better and cheaper than a lot of the overpriced, crappy food in Europe.

It's a fake image of a medieval utopia that never existed even in fairy tales.

That's quite a high horse you rode in on.

This. Exactly this. No sophisticated tokenization. No interesting architecture using attention. And the author is completely clueless about overfitting... and even cross entropy loss. He could have gotten better results just using a bag of words approach.

But this ends up on frontpage anyway. Welcome to HN.

Were you born after the year 2000 or something? I am talking about the original Doom, not Doom for millennials.

John Romero created the original Doom, not 2016.

You are wrong. Dead wrong. Doom does not have a big dumb story, dialogue, cut scenes etc. It has great gameplay, pure and true.

I want gameplay, I could care less about some big dumb story.

There is also a "team".

So it was for the best.

Exactly. GPUs are necessary now, but did not originally herald the deep learning revolution.

Huh? Carriers are intended to be mobile bases for launching aircraft that go on to stage air strikes. While they are armed, they aren't intended for attacking other ships.

And what size of payload would a drone need to take down a carrier? That is just silly.

VGG was a long time after the first deep learning revolution.

Seen it all before. In practice, this doesn't work. You end up way too dependent on a superstar who is bound to burn out, and will take the whole team down with him.

That kind of thinking is along the lines of "just always pass Michael Jordan the ball". But that's not how healthy teams function. And it wasn't how Phil Jackson managed his team.

And you can credit John Carmack all you want. But his peak was when he worked alongside John Romero. Even Carmack himself has stated this. And Michael Abrash taught him the theory of how to do 3D rendering properly.

There is no "I" in "team".

It is a good manager's responsibility to maintain the health and longevity of his team. Obviously, that wasn't the case here, and the burnout superstar even left. Not surprising.

"As a user... The $100/year dev license kinda sucks if you just want to hack around and distribute code"

As a user, what does $100 a year even accomplish? It is nothing more than a racket.

This should be the top comment in the thread. This kind of work schedule is not healthy or sustainable. Long term success is like running a marathon.

Except that's "currently". NNs can and were trained on CPUs originally.

the resurgence of ML ... because video games - of all things - funded the maturation of a whole new category of massively-parallel chips

Fake news. It has far more to do with the rise of distributed computing than the existence of GPUs.

That doesn't mean startups are meaningful.

No one said working for startups is meaningful.

Not with pubsub. You could use it as shared memory store for player states though, I guess.

3Blue1Brown is not very good. I have found Udemy courses way more valuable. Even Stanford or Coursera lectures can be better.

I think 3Blue1Brown is a good example of things being trivialized to the point where you're tricked into thinking you learned something new, when you actually didn't learn much at all.

that for some people start ups are a fashion or a "thing to do"

This is true, and I was expecting the same sentiment from the post but it was not there.

And as far as startups being a "thing to do". As opposed to doing what else? Working for a big company? What pray tell would be better and more meaningful? It probably wouldn't be in tech.

Use Redis pubsub instead.

Correlation is not causation. Everyone watched Disney movies as a kid.

There's no study that supports post 2000 content does anything to improve things.

There is also Google OR tools.

https://developers.google.com/optimization