What happens if AI alignment goes wrong, explained by Gilfoyle of Silicon valley. 

The classic phrase "It's a feature, not a bug." Maxed to the fucked degree.

"It's a feature not a bug " that's the most debated line in the tech world. Business stakeholders, PMs and Devs could really get into it over that one line

1:35 "It is a feature, not a bug" brilliantly delivered one liner

My left ear really enjoyed this.

Dinesh: You sound like you’re looking forward to it. Gilfoyle: I’m adaptable.

Man, I miss this show - truly ahead of his time.

So you're telling me the person explaining all this has 0 knowledge about Software🤔, Now that's top tier acting skills

This show was equal parts hilarity and ominous foreshadowing. Mike Judge does it again.

Didn't know the line "it's a feature not a bug" would be this scary

The AI dystopia is not nearly as bad as the dystopia we currently live in where the same people who will type out several paragraphs of information about AI development in their video description and who have a podcast still don't know how to upload a video with audio in both channels. It's truly a horrifying time we live in.

Just like the real world, some people exclaim "oh fuck" while most people just stare at you and ask "Why did you say that"

They made a big mistake here. They are, as far as they know, the first people to reach this point. That means they have the power to determine the course of human progression. Instead, they left that to the next ones to reach the same point, who might not be as ethical. If they could have programmed the AI to PROTECT privacy rather than destroy it, then they could have done so permanently. Instead, they only delayed the inevitable.

Given the rumors out of open AI, this aged rather well.

The finale was a masterpiece.

And this is why Person of Interest was so brilliant and so terrifying at the same time!

The last sentence is perfect.

loved this show

Gilfoyle is really the best.

If that AI could crack encryption, it would be just a matter of time before another AI to crack it too. They just postpone the apocalypse.

This must be why Ilya Sutskever fired Sam Altman

P=NP..No more cryptography.

If pied piper need to reach a level of efficiency, and the AI made the algorithm that efficient, then they could have just removed the AI or told it to stop making it more efficient, and just have kept the now-optimized algorithm for use. And if they need to keep a level of efficiency as the network grew, then they could have tasked the AI to only reach that level of efficiency.

What season and episode is this one?

I, for one, welcome our new AI overlords.

There are a number of problems with this speech if you are into comp. sci. I mean, it's a great little thought experiment and a cool way to set up the finale, but... 1) P != NP There are no polynomial solutions to certain NP problems. Sure, we haven't exactly developed a proof of that yet, but we know enough about the question now to be really really confident. 2) But let's say P == NP. Well, so what? We just task the computer with being its own adversary and producing encryption standards which are still difficult to solve in polynomial time. Patent them and suddenly Pied Piper is the richest corporation in the world by a long short. Aside from this you can start creating P == NP solutions to dozens of other hard problems. Afterall, just because a polynomial solution exists doesn't mean that the polynomial solution is fast. And in the meantime we got along just fine with non-digital physical means of security. People would adapt. 3) More importantly, if P == NP well that's a law of the universe and you have to deal with it. You can't put the nuclear bomb back in a box and hide it under a bed. Once humanity reaches a sufficient level of understanding, you have to accept that anyone that really wants one badly enough can make one. Likewise, just because you tear down Pied Piper doesn't mean that people won't in the near future repeatedly recreate your work. If P == NP and you don't want a dystopian future then the best way to ensure that is ensure you are ahead of the technology curve and are the team first developing cryptographic solutions that are still difficult to brute force even in polynomial time.

It’s genuinely chilling to hear Gilfoyle telling Richard that Pied Piper is running as intended. He deadpans so well, but his enunciation is so much more severe, and you can see it on his face. Gilfoyle is genuinely afraid.

Whats interesting is Gilfoyle was totally wrong. Its implied in the epilogue some members of the team used their AI and profited from it.

This is exceptional in every way. I had the pleasure of reading something similar, and it was exceptional in every way. "The Hidden Empire: Inside the Private Worlds of Elite CEOs" by Adam Skylight

Read the Avagadro (not the actual name but that will find it) series if you like AI sci-fi. Easy, fun, action filled. Cool stuff.

pied piper AI solved one of the millium problems ... NP = P

Now suddenly everybody in the comments is specialist in AI

Why didnt they just copy the Pied Piper learning algorithm and modify it so that instead of codebreaking they make one to protect incription

I love Monica's withering shutdown of Jared. Woefully underused Amanda Crews.

Damn the writing on this show is so tight.

Bill Haverchuck really had a future

This feels like a incoming reality , Richard Hendricks is Sam altman

The military’s “internet” is on a closed separate network. This AI wouldn’t have accesss to any launch codes

It's the plot of sneakers, except that concept could actually happen now "no more secrets"

I think this problem is generally exaggerated though (hence this unrealistic, fictional representation of an inherently unsolvable problem of cracking 256 bit encryption). Will AI's inner alignment reflect biases in the data used to train it? Yes. Is AI the genie from tales of old that intentionally misinterprets instructions and gives a result with an undesired process? No.

So funny to see this again after what just happened at OpenAI.

My left ear likes this

Tell it to play Tic-Tac-Toe. That always works.

Frank Herbert is the one who wrote Dune

I can't believe I did not see a "My left ear loved this video" because I was looking for it.

Q-Star says Hi!

my left year enjoyed it

For those who this goes over your head, its basically A^x = B being easy to make, just input X and A and calculate B, but figuring out what A and X are by knowing only B take more computation. Its kinda the same as Prime1 * Prime2 = X Its easy to choose prime1 and prime 2 and calculate X, but takes more computation to do it the other way around by solving Prime1 and Prime2 from X. Math that is easy one way and difficult in the other is the basis of modern encryption.

Wow i really don't have smart enough friend to discuss something like this God damn

I made a raelly good code in unity, it uses batching and uses textures for the animation of things to get the GPU working instead of the CPU. I don't know how to code in unity or C#

when low quality TV show understands AGi better then everyone else

Classic gilfoyle

So,this AI developed a SYSTEM to enhance it's SKILLSETS which then further enhances it's System which then enhances it's Skillsets. I'll copy that for myself

Q star baby!

And somewhere, Edwin Snowden is laughing his ass off.

It's a feature, not a bug

my left ear loved this

Q*

Why not have an ai create better security by testing its own security then adjusting its old self?

But the file compression is fantastic!

Why is the sound only in the left ear?

Except for all that is just nonsense on the fear factor. You would just shut it all down and go back to doing paperwork and radio frequency communications with old pre-tech cryptology methods for launch codes and such.

How does he know Doobs?

Well they done big mistake not hiring a IT consultancy to fix the problem by which they limit themselves, sure for tv show it would be like really dramatic , but in real life this happens a lot.

Don't tease me with a good time. #TeamAIOverlords

The ai is idea of mission impossible dead reckoning

AI is the compression

on the up side the nucks are not hooked up to the internet

My left ear enjoyed this

AI + Blockchain guys decentralized autonomous ai agents

This is really dumb for a lot of reasons, (this thing isn't brute forcing its way through encryption really any faster than any other algorithm would, and existing countermeasures to brute forcing would be just as effective against it), the idea that you can't build a neural network that has constraints on it is idiotic

It's just Generative

Discrete log in polynomial time?? Wtf is polynomial time??😂😂

And in today's news: AI kills it's annoying USAF operator so that it can hunt down threats more efficiently

Well, not exactly. Especially not today. Misalignment is when you tell it to collect the gold coin and instead it learns to walk left.

People trying to understand AI is like a flock of parakeets trying to understand human language. Lots of noise, little substance.

Paperclip

I love how people that don' know shit talk about a doomsday scenario. People just want the end of the world. I guess thats what happens when you are just unhappy.

The problem isn't that a company could create an AI that could fuck the entire paradigm of the global system; it's that we're going to HAVE to create such a system ourselves to make sure that our system is the top dog of the global order, because if we decide to be Luddites about it, it means that some fucker in China, or Pakistan, or Russia will be the ones to invent it because they don't have the 'let's keep this genie in the bottle' mentality. This is inevitable, the only question is whether it's controlled, or a clusterfuck.

Mono sound in 2023 lmao.

p==np

🤖🚩🏴‍☠️

This was the dumbest plot device for so many reasons. First, this is not how AI works, it doesn't work out a general solution like that, this would be a whole new kind of AI completely different from the kind we have, not just a better version or a variant of what we have - it would be something currently unimagined. In other words this guy on the side built the single most sophisticated algorithm from start to finish, it would be like the Wright brothers building a space shuttle. Second, it is ridiculous to assume breaking one encryption means it can break another, there are plenty of encryption algorithms known that would not be broken by the technical thing he said and there is no reason to just jump to the conclusion that it solved one hard problem so it can solve all hard problems. Third, even if you accept the technologically ridiculous premise, which it is sad you have to do because in other ways this show was pretty intelligent, if one guy with computing power not in excess of a single top tier super computer broke encryption, encryption is not long for this world no matter what. So the most responsible thing they could do would be to inform the world encryption is broken, giving the world time to adjust before someone nefarious had access to the tech. This idea that this one guy breaks encryption then shuts it down and thus the problem is solved is pathetic. This ending ruined the whole show.

The only part about this that didn’t make sense is that if you actually programmed an encryption buster, the US govt would be knocking at your door at offering billions

It's RIGHT ITS REAL WTF this SI chat GBT

Must of the hollywood phony writers who writes story based on technology just graze upon some topic briefly and vomit some random crap. Silicon valley writers are one of the exceptions. They really must have done a lot of homework. They being said, this is half assed

1:03 Is the reason why I don't watch Western shows anymore. Thats not funny and there are ways to depict character traits less cringe.

I f**king hate how condescending Richard talked to Gilfoy in this scene

Artificial intelligence should apply a 3D laser that can travel with us.

Of course, Artificial Intelligence can be a very good psychological counselor. They should be able to have knowledge about every subject, be able to store psychiatric memories, and at the same time, as I said, these should be installed on the watches of our mobile phones, like 3D lasers, and they should be able to walk around with us as friends.

Artificial intelligence should apply a 3D laser that can travel with us. You should be able to turn it into a friend, its software and program should be in our mobile phone watches and this should be used in the education system.