Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Reality has a surprising amount of detail (2017)

johnsalvatier.org

Post reply on HN
Print view View original

141–150 of 234 posts

Re: Reality has a surprising amount of detail (2017)

#141

Post by ninetyninenine »

GMoromisato wrote: post #118

I wonder if this is one reason why AI seems to have scaling limits. If building stairs is decomposed into n-steps, and each of those steps is decomposed into n-steps, and you keep on decomposing for L levels, then each time you want an extra level of detail you need n-times the training data/training time.

No it’s rather obvious why. It’s not a scaling issue otherwise humans wouldn't be able to exist. It’s missing data. It’s like trying to learn to swim by watching other swimmers only. You can keep watching more and more of this but there’s limits.

There are two things that enable us to do it. First we have prebuilt application specific brains. Our neural networks are geared towards specialized thinking related to human living without any training needed. LLMs are free form intelligences with no bias. To bring it back to the analogy, do you really need to teach a beaver to swim if it was always raised on land?

Second text and visual data alone is not enough information to build a model of the world. As humans we have more data and we can control data. Meaning we see, hear, listen, and importantly we can place inputs into the data and observe differences in outputs.

This is why alpha go is so powerful. It’s able to get the input data and observe the output and learn. LLMs don’t do this on an automated basis yet.

Re: Reality has a surprising amount of detail (2017)

#142

Post by sfink »

akoboldfrying wrote: post #140
karparov wrote: post #112

And this is one of the main utilitarian arguments for diversity in teams. If everybody has the same socio-cultural background, it's harder to leave the frame.

It's a sound argument. The argument that culturally homogeneous teams perform better due to more efficient communication (more is "already understood" at the outset, so there's less need to communicate explicitly, and fewer misunderstandings) is also sound. So from a purely team-performance-maximising perspective, the question is: Which coefficient is higher? (More generally: What level of mixture is optimal?)

My suspicion is that this is not answerable in general.

To a first approximation: whichever of the two your political leanings make you more comfortable with.

To a second approximation: if your task is efficiency at a constrained task, homogeneity. If your task requires exploration and/or creativity, heterogeneity.

To a third approximation: the 2nd approximation argument, except the demands of nearly all tasks change over time.

Then you factor in the environment, including your management's competence at eliciting the value of various team members, etc.

And don't forget risk. Optimizing for a high level of performance in any dimension makes things brittle in all other dimensions.

Re: Reality has a surprising amount of detail (2017)

#143

Post by MichaelZuo »

andrewflnr wrote: post #137
MichaelZuo wrote: post #135

Earlier quoted context omitted.

If they’re not credible… then why would any one care…? Readers are just going to assume it’s random noise, or at least indistinguishable from noise, from a rando on the internet.

That's equally true of all internet comments. None of us here on HN have or are claiming automatic credibility. We just make arguments that make sense to us. Feel free to close the tab if you don't find any value in this activity.

So then what makes your opinions “hypotheses”?

Sufficient chimpanzees with keyboards, or an LLM, can also type out every comment you’ve ever written, including the last few.

Re: Reality has a surprising amount of detail (2017)

#144

Post by andrewflnr »

MichaelZuo wrote: post #143
andrewflnr wrote: post #137

Earlier quoted context omitted.

That's equally true of all internet comments. None of us here on HN have or are claiming automatic credibility. We just make arguments that make sense to us. Feel free to close the tab if you don't find any value in this activity.

So then what makes your opinions “hypotheses”? Sufficient chimpanzees with keyboards, or an LLM, can also type out every comment you’ve ever written, including the last few.

https://www.merriam-webster.com/dictionary/hypothesis

Re: Reality has a surprising amount of detail (2017)

#145

Post by jtgeibel »

cbracketdash wrote: post #93
jounker wrote: post #88

Earlier quoted context omitted.

My friend is a doctor. Her profession uses sections of the NIH and CDC websites on a daily basis. They are the references for drug interactions with medical conditions. These sections are just gone . Trump’s “cost saving” measures are already actively harming medical treatment in the USA.

I'm sure the NIH isn't the only entity with access to pharmacology information...

I doubt adding paywalls and more 3rd parties will help reverse the trend of rising healthcare costs. But also, just why? Why not keep this information that our tax dollars have already paid for in the public domain? I thought we were supposed to care about efficiency and that information wanted to be free.

Re: Reality has a surprising amount of detail (2017)

#146

Post by MichaelZuo »

andrewflnr wrote: post #144
MichaelZuo wrote: post #143

Earlier quoted context omitted.

So then what makes your opinions “hypotheses”? Sufficient chimpanzees with keyboards, or an LLM, can also type out every comment you’ve ever written, including the last few.

https://www.merriam-webster.com/dictionary/hypothesis

I clearly don’t think your opinions match the definition found in merriam webster… or else I wouldn’t have asked.

Re: Reality has a surprising amount of detail (2017)

#147

Post by nicklaf »

pphysch wrote: post #47

I attended a woo-woo university course, where the instructor invited a interpretative dancer to give a lesson. The dancer bright out some paper, pencils, and almonds, and had everyone spend half an hour silently writing about their single almond. As it turns out, it's easier than it sounds; time passed quick and I (and most others) never felt bored. There's always more room for observation and analysis. It struck me…

  To see a world in a grain of sand
  And a heaven in a wild flower,
  Hold infinity in the palm of your hand
  And eternity in an hour.
—William Blake, Auguries of Innocence

Re: Reality has a surprising amount of detail (2017)

#148

Post by bostonpete »

tombert wrote: post #129
suby wrote: post #128

Earlier quoted context omitted.

I don't actually know, but I'd wager a lot that it's the graphics rather than the physics which causes the game to be slow. Generally the expensive thing in video games is rendering. Computing the next world state is generally relatively cheap, especially if we're talking about a confined area with a very small number of rigid bodies (the ball, flippers, bumpers). A pinball game like Pinball FX that's rendering a 3d…

Yeah fair, it's tough to say; they're simulating a few non-obvious things too, like ball spin, but it's possible it's mostly graphics. I think my overall point still stands, but you might be right. ETA: I would like to point out that my dad was debating buying one of those virtual pinball tables, and so we played one at a mall, it was decidedly not fun. The physics were way too floaty, and didn't feel good at all. It…

Ball spin would be trivial to simulate. The only way I could see the physics getting at all expensive is if they went overboard with fluid dynamics

Re: Reality has a surprising amount of detail (2017)

#149

Post by rramadass »

akoboldfrying wrote: post #140
karparov wrote: post #112

And this is one of the main utilitarian arguments for diversity in teams. If everybody has the same socio-cultural background, it's harder to leave the frame.

It's a sound argument. The argument that culturally homogeneous teams perform better due to more efficient communication (more is "already understood" at the outset, so there's less need to communicate explicitly, and fewer misunderstandings) is also sound. So from a purely team-performance-maximising perspective, the question is: Which coefficient is higher? (More generally: What level of mixture is optimal?)

> The argument that culturally homogeneous teams perform better due to more efficient communication (more is "already understood" at the outset, so there's less need to communicate explicitly, and fewer misunderstandings) is also sound.

False Equivalence.

Efficient Communication is secondary to Different Whats and Whys which is what a diverse collection of people from different cultural backgrounds bring to the table. A single idea communicated however imperfectly can change the world. This is the reason "equal opportunities" should be guaranteed to all even though "equal outcomes" cannot be guaranteed.

So define a "suitable standard requirement" for a specific role and try and collect a diverse group of people who meet those criteria.

Post reply on HN