randartie

“ A video is just a sequence of frames.”

Except they showed their AI the 3 most important frames (rock paper scissors) and not the hundreds in between.

What is our consciousness other than a set of physical chemical reactions?

You can task the human-level AGI to optimize itself. You can create an army of copies of itself to work towards that. It's self fulfilling.

The way I view this as a net-gain is that AGI agents will be able to have innovation on improving themselves (exponentially?) and the whole world. They very well may be the path to a utopia.

Task 1 for AGI: Optimize your own system so you don't cost 50 million dollars to train.

" If he gets bored/throw in towel, he can just pay back with his money and move on."

"Just to train a large model once will eat up 20% of this money."

Aren't these 2 statements conflicting?

If I had to speculate based off my experience at other FANG companies, the videos still exist in the Youtube infrastructure somewhere to allow for the possibility of a human intervention to restore them (For example, to allow for recovering from a system gone haywire or just to give time for manually reversed decisions).

That said, I'd also guess there's a countdown timer before they get deleted permanently.

I hope you can get a human from YouTube to talk to you.

It may benefit you to post a link to the youtube channel so someone can proactively investigate. Lots of google employees read this forum.

Do you mean game development in general? Or the author of the blog?

I believe Bevy (rust game engine) does leverage multi-threading for running game code (most 'data oriented' frameworks do). The most popular older engines like Unity do not (unless using the experimental DOTS framework).

So, the phrase "a tremendous amount of game code doesn't benefit from multithreading" is actually true since most game code is in unreal/unity, but it's not a hard constraint for all game engines.

If an app from google uses 1000 open source projects (probably more in practice), how do you distribute the x%?

This is mostly powered by amazon cloud9, which is really cool in itself. https://aws.amazon.com/cloud9/

It is a facebook-controlled private currency

Can you clarify? Isn't the currency going to be owned by a consortium of various global businesses [1]? Are you unaware of that?

[1]https://libra.org/en-US/association/#founding_members

Am I the only person who is impressed with this person's debugging skill?

”But Watson won’t change its conclusions based on just four patients. To solve this problem, the Sloan Kettering experts created “synthetic cases” that Watson could learn from, essentially make-believe patients with certain demographic profiles and cancer characteristics.”

Is this standard practice in machine learning? This sounds more like regular programming to get exactly the outcome you want.

This sure is getting a fair bit of downvotes. I mostly agree with you though. Having a single identity reduces trolling, it's as simple as looking at Facebook vs Twitter. People are less likely to message nasty things to others if their real life reputation is on the line, I take that as fact.

I routinely disable my adblock for certain news websites because they tell me to, otherwise I can't see their content. Websites asking for 'opt-in' does work. It works on me all the time.

0) Initialize set X to contain every word.

1) Y = set of words in every definition of the words in set X

2) X = Y - X (all words in Y that are not in X)

3) Repeat from 1 if the set of words in X has changed

Does that reduce all words down to the actual minimal set of words required to define other ones? Since you can build upwards from the resulting set X to get the original set of words.

Also, this reminds me of the knapsack problem a little bit (for example what is the minimum set of coins required to be able to make $X).

With TenCent having the pull that they do in China, I wonder if this could mean Snapchat possibly being unblocked in China soon.

The cynical part of me thinks that when a big purchase like this happens, the buyer knows something we don’t.

Then again, they bought 5% of Tesla as well this year.

Link to facebook group? A search for 'Starcraft AI' doesn't bring up any results on facebook for me.

Half in what sense?

Needing this amount of resources implies that Snap is expecting huge growth. This sounds like a really bad move on their part and they should have committed to building out their own infrastructure on 'bare metal' over the next 5 years instead.

If you read their S-1, they list a dependence on Google cloud as one of their big risk factors. Yet they then go ahead and make this commitment instead of working towards eliminating it.

There's so many advantages to owning your stack and if Snap thinks that it's going to need 2 billion dollars to pay for cloud infra, they're at the scale where it makes sense to build your own infra. Just look at Facebook, they're able to create tailor made data-centers that fit precisely what they need. The success of Snap relies on huge scale on the consumer side, if they want to scale their infra to support that 5 years down the line then this sounds like a poor move since they will either need to play catch-up later on or prepare to pay serious dough to Google.

Paying for cloud services seems like a great idea when you are not able to predict your needs in the coming years, given a deal like this I don't think that's the case.

Kind of ridiculous IMO that Google does not extend offers before leaving the internship. In cases like this, it really does not make sense.

Wouldn't the type of prototype created in origami be past the index card stage, and should be the time to think of things like that?

Seems like this story is being received pretty well here.

Though, in my opinion I think it's kind of insulting to be asked something like "What did you learn?". The question isn't really necessary. You know that you screwed up.

That kind of dynamic between an engineer and a team lead is off-putting to me.

I think the proper way for a team lead to handle it is to instead work with the engineer to help find ways to eliminate the human error by implementing tooling or processes. The conversation should go something like this:

Lead: "I wonder how we can make sure none of us break XYZ widget again"

Engineer: "We can build out ABC and run that, also generally just test better before pushing to prod".

Lead: "Cool, do you want to go ahead and take care of that?"

From the demo filmed today vs that demo, it feels like they've made a lot of progress

How can they possibly staff a 5000 person ai division. Sounds to me like only a small-ish subset will be talented ai researchers.

If they can make them at a cost of $0, sure

Okay, but there's plenty of highly liquid things to trade. For example the SPY has a spread of 1 penny, same with SPY options.

You mention a potential strategy for manipulating the market like pushing prices to hit stops, well if these strategies are common then it seems there exists a potential strategy to take advantage of this.

It seems like so long as the market's next tick price is not completely uncorrelated from its previous tick price, then there will always be a way to 'beat' the market. The market price is not completely uncorrelated (as you mentioned, there's price manipulation driving it to certain prices throughout the day).

You can't just flip all your decisions because that would only work if your decisions are inversely correlated to the right decision. Losing traders are usually completely uncorrelated.

What's stopping ISPs from slowing down other speedtests at the same rate as netflix to eliminate your ability to have a 'control'?

By your thoughts, all software should have been updated in a few days to be bug free.

Anyone who has used this app knows it is extremely crashy. It's not supposed to be bug-free, but releasing an app this crashy is fairly sloppy.