Live data from Hacker News

We accidentally solved robotics by watching 1M hours of YouTube

ksagar.bearblog.dev

171–180 of 183 posts

Re: We accidentally solved robotics by watching 1M hours of YouTube

#171
post #148

Earlier quoted context omitted.

Indeed, the robotics edge-case problem space complexity balloons far faster than most assume. Physics informed training is a real methodology (simple introduction to the subject: https://www.youtube.com/@Eigensteve/videos ). However, the slop article is 80% nonsense. =3

Just the hand, you have 50 things you just can’t do unless you have certain feel. Handling glass? Oh it’s greasy, now your rubber grip is screwed, now go wash it off, and dry it to start again.

Indeed, there is a whole category of Adaptive Compliant Grippers for specific use-cases like handling delicate brittle objects.

Things that randomly change shape or appearance are also very difficult to interact with safely. The force sensing platform from Universal Robots is safer for users, but it has limitations like all platforms. =3

Re: We accidentally solved robotics by watching 1M hours of YouTube

#172
It seems that in order for robotics and automation to work properly, AI models including LLMs, YOLO, RL and others need helps from their distant cousins namely logic, optimization and constraint programming that can be attributed as intelligent automation or namely IA [1],[2],[3],[4].

[1] Logic, Optimization, and Constraint Programming: A Fruitful Collaboration - John Hooker - CMU (2023) [video]:

https://www.youtube.com/live/TknN8fCQvRk

[2] "We Really Don't Know How to Compute!" - Gerald Sussman - MIT (2011) [video]:

https://youtube.com/watch?v=HB5TrK7A4pI

[3] Google OR-Tools:

https://developers.google.com/optimization

[4] MiniZinc:

https://www.minizinc.org/

Re: We accidentally solved robotics by watching 1M hours of YouTube

#173

Earlier quoted context omitted.

What I describe is an unsupervised system. What you say ("interventional") sounds like it's human-supervised. But maybe I'm interpreting it in the wrong way, so please correct me if so.

By "intervention" I mean interacting with the environment. Purpose a hypothesis, test, modify, test. You can frame RL this way though RL usually generates hypotheses that are far too naïve. This looks like a good brief overview (I only skimmed it but wanted to give you more than "lol, google it") http://smithamilli.com/blog/causal-ladder/

Yes, you need to let the robot play (interact with the environment) to learn the vision-versus-touch correlations, but you can do so in an unsupervised way (as long as you choose the environment wisely).

Re: We accidentally solved robotics by watching 1M hours of YouTube

#174

Earlier quoted context omitted.

Crimes generally don't kill the criminal. It's the reaction by authorities that kills (perceived) criminals.

This is true. In general, the harm done by crime is directed outwards from the perpetrator, not inwards to the perpetrator. In fact, the behaviors that only cause self-harm that we criminalize are relatively few.

So however you want to twist it, he was killed by the government.

Re: We accidentally solved robotics by watching 1M hours of YouTube

#175

This is interesting for generalized problems ( "make me a sandwich" ) but not useful for most real world functions ( "perform x within y space at z cost/speed" ). I think the number of people on the humanoid bandwagon trying to implement generalized applications is staggering right now. The physics tells you they will never be as fast as purpose-built devices, nor as small, nor as cheap. That's not to say there's zer…

The value is in the generalisation. For a single example, in any factory watch how humans are added as ad-hoc machines wherever a problem occurs. Machine N outputting faster than machine N+1 can accept? Have a human stack, and destack, the product between them. No matter the size, shape, it within reason the weight of the product. But most importantly: the process can begin within seconds of the problem occurring. No…

Your assumption set: conventional factory space, idle humans, traditional management, ad-hoc process with skilled managers. This is similar to the "job shop" mentality in (dying) manufacturing. You additionally assume general purpose magic hardware that can usefully do anything.

Reality: Most value is in shrinking things, excluding humans, automating management, carefully designed process, and specialist hardware that does a subset of things very well. Relying on human(oid)s is a sure-fire way to suck.

Re: We accidentally solved robotics by watching 1M hours of YouTube

#176

Earlier quoted context omitted.

This is true. In general, the harm done by crime is directed outwards from the perpetrator, not inwards to the perpetrator. In fact, the behaviors that only cause self-harm that we criminalize are relatively few.

So however you want to twist it, he was killed by the government.

Sorry, I don't see how you arrive there from the fact-pattern. He wasn't a criminal because he never had a trial. He killed himself before he even had a hearing on whether the prosecution's evidence was admissible, much less his opportunity to either prove his innocence or argue the acts he undertook shouldn't by rights be a crime at all.

What should the government (executive or judicial) have done differently to balance the needs of the accused vs. the needs of the enforcement and adjudication of the law here?

Re: We accidentally solved robotics by watching 1M hours of YouTube

#177

Earlier quoted context omitted.

The value is in the generalisation. For a single example, in any factory watch how humans are added as ad-hoc machines wherever a problem occurs. Machine N outputting faster than machine N+1 can accept? Have a human stack, and destack, the product between them. No matter the size, shape, it within reason the weight of the product. But most importantly: the process can begin within seconds of the problem occurring. No…

Your assumption set: conventional factory space, idle humans, traditional management, ad-hoc process with skilled managers. This is similar to the "job shop" mentality in (dying) manufacturing. You additionally assume general purpose magic hardware that can usefully do anything. Reality: Most value is in shrinking things, excluding humans, automating management, carefully designed process, and specialist hardware tha…

Correct, I'm talking about the 98% of factories in the world today and in the near future. Obviously the far future will see changes in manufacturing, just as manufacturing has seen changeds every decade since we've been manufacturing things at scale.

Re: We accidentally solved robotics by watching 1M hours of YouTube

#178

Earlier quoted context omitted.

So however you want to twist it, he was killed by the government.

Sorry, I don't see how you arrive there from the fact-pattern. He wasn't a criminal because he never had a trial. He killed himself before he even had a hearing on whether the prosecution's evidence was admissible, much less his opportunity to either prove his innocence or argue the acts he undertook shouldn't by rights be a crime at all. What should the government (executive or judicial) have done differently to bal…

The government killed him by threatening insane punishment for something that is practically harmless, and relevant to the original point, is done without a second thought now by the bigcorps to feed their AIs

Re: We accidentally solved robotics by watching 1M hours of YouTube

#179

Earlier quoted context omitted.

Sorry, I don't see how you arrive there from the fact-pattern. He wasn't a criminal because he never had a trial. He killed himself before he even had a hearing on whether the prosecution's evidence was admissible, much less his opportunity to either prove his innocence or argue the acts he undertook shouldn't by rights be a crime at all. What should the government (executive or judicial) have done differently to bal…

The government killed him by threatening insane punishment for something that is practically harmless, and relevant to the original point, is done without a second thought now by the bigcorps to feed their AIs

Prosecutors do that all the time. Basically nobody dies of it. I'd humbly propose there were unfortunate mitigating circumstances in Mr Swartz's situation that made it unusual. When a person with AIDS dies, did the AIDS kill them or the pneumonia a regular body would have fought off? When a person with deep mental illness commits suicide, did the circumstances of their life kill them or did they succumb to a deep mental illness?

Perhaps we could craft a way to hold people with mental health issue to the same standards we are all held to while simultaneously being more sensitive to their needs. But in general, his story is an unfortunate tragedy of a sick person who took their own life under a stress that doesn't kill most other people, and we adjust the way we prosecute crime at our own peril. It is, as I said elsewhere, the mother of all Chesterton's Fences. Which is not to say it cannot or should not be improved! Only that it be done with great care.

And to be completely clear: Swartz ripped content via back-dooring a secured network physically, in a closet, and (it is alleged) planned to dump that content in public. We'll never really know since he (or his illness) denied himself his day in court, and that's a tragedy; he may have successfully defended himself, or could have been a living example of persevering anyway like Mitnick instead of a martyr. Companies using their authorized accounts to scrape Google are likely at most guilty of a TOS violation and Google may choose to cut their accounts, but it's very hard to make a case that the Google API saying, over and over again, "Yes you may view that video" constitutes either unauthorized access or exceeding the bounds of access under 9-48.000.

It's hard to comment on whether Swartz violated the CFAA. Since he wasn't tried, we'll never really know. He exited life before justice could happen one way or the other.

Re: We accidentally solved robotics by watching 1M hours of YouTube

#180

Earlier quoted context omitted.

Many people get (for example) pneumonia and recover. Some people get pneumonia and die. The people who died of pneumonia died because of pneumonia. The fact that other people survived it doesn't mean that they didn't die of it. Saying that we should not work on cures for pneumonia because it's a Chesterton Fence is obviously, blatantly, illogical. Saying that we should change the system so that government officials w…

Pneumonia doesn't have any societal benefit. The process by which we decide if the law was broken and punishment necessary has obvious benefit. If you mean we should seek a cure for dangerous suicidal depression, I agree. But you surely are not suggesting that, for example, has Swartz been accused of embezzlement that the state drop out finish the charges purely because he's a suicide risk; how would that be just to…

We evolved with pneumonia for some reason. It could easily be a Chesterton Fence. We don't treat this as one because we don't want people to die of it.

I agree that a system of laws has benefit to society. However the system we've worked out for making such laws is clearly being warped and twisted to serve one small section of society at the expense of everyone else.

A clear case being the comment that started this conversation - Swartz was hounded to death for doing the exact same thing that AI companies are doing and they're facing zero punishment. AI executives are not being dragged from their offices by burly policemen and thrown into cells, yet they have done the exact same thing that Swartz did to merit that behaviour. It's not unreasonable to question the societal benefit of this system.

And we totally should say that people died of depression, or financial stress, or legal persecution, or whatever. Most people have suicidal ideation at some point in their lives, that's not unusual. Being hassled to the point where you go through with it is definitely violence. Classing this as "mental illness" and therefore a personality defect is a form of victim blaming.

Post reply on HN