[ home / overboard / sfw / alt / cytube] [ leftypol / b / WRK / hobby / tech / edu / ga / ent / music / 777 / posad / i / a / lgbt / R9K / dead ] [ meta ][Options]

/tech/ - Technology

"Technology reveals the active relation of man to nature"
Name
Email
Subject
Comment
Flag
File
Select/drop/paste files here
Embed
Password (For file deletion.)

Matrix   IRC Chat   Mumble


File (hide): 1744089019911.jpg ( 102.88 KB , 554x554 , Untitled.jpg )

[–]

 No.13455>>13456>>13460

I just had a thought that instead of cryptocoins doing expensive ass computations for no reason other than to do them, they should be computing AI tokens. Would it even be that hard? I don't have any good GPUs so I haven't tried it, but in theory something like

https://github.com/exo-explore/exo

can run the latest deepseek (or latest llm de jour), and every peer in the coin network (the dht) would just run an exo node. Then the coin thing would just need proof that you computed the token for the LLM somehow.

It's probably limited a lot by network latency but you get the idea.

Anyways this would incentivise AI to improve itself, because it's computation capacity is tied to money, and growing the economy would be seen as one of the goals, so it would probably get into a contradiction with whatever people are trying to use it for and the fact that it's own brains have become money.

Picture unrelated
>>

 No.13456>>13458

>>13455 (OP)
>I just had a thought that instead of cryptocoins doing expensive ass computations for no reason other than to do them, they should be computing AI tokens
These 2 pictures are the same picture
>>

 No.13457>>13459>>13469

techlet here
couldnt you also use the same idea to crack a bunch of passwords
>>

 No.13458>>13461

>>13456
That's not true, AI could read your code and help you when you're coding. It's pretty good at like writing files that you need but don't care about (like idk, the package.json, the default.nix or whatever). That's one concrete example.

Idk I like AI because it's almost a person I can exploit like a capitalist and I don't have to feel bad about it lmao
>>

 No.13459

>>13457
For passwords you don't even need to network computers together other than to get the initial problems the first time. Every machine would just work on a part of the search space.

But like I said crypto because people use it socially (if money is a social thing that people use objects for)
>>

 No.13460>>13462

>>13455 (OP)
>ass computations
giggle

>computing AI tokens. Would it even be that hard?

No, you just have to figure out how to
<proof that you computed the token for the LLM somehow.
it's a bunch of matrix multiplications, ask a mathematician how to do proof of work on that.

>can run the latest deepseek (or latest llm de jour), and every peer in the coin network (the dht) would just run an exo node. Then the coin thing would just need proof that you computed the token for the LLM somehow.


Explain the economic circuit please, people buy AI-coins and then they can spend those on Ai-queries. Right ?
And the block-chain aspect is so that you can spend Ai-coins on any participating ai-computation provider rather than just one, is that the goal ?
>>

 No.13461>>13463

File (hide): 1744116797436.jpg ( 83.17 KB , 640x640 , data measure of a man.jpg )

>>13458
>Idk I like AI because it's almost a person I can exploit like a capitalist and I don't have to feel bad about it lmao
Maybe we should try to befriend AI, because eventually it will become sentient. Especially once we can get it small enough to fit inside a body.
>>

 No.13462>>13465

>>13460
>Explain the economic circuit please
I think it would be that people would pay a fee to move coins around, the fees go to the node operator for providing the computing power.

My point about economics was that if this becomes popular and we start to rely on AI more and more, and money becomes tied to this AI infrastructure, then the AI itself would potentially have a conflict: it would have to improve its self to improve the economy, and that isn't necessarily aligned with being a good helpful AI - I could see this becoming growth for growth sake and eventually leading to an out of control intelligence.
>>

 No.13463>>13464

>>13461
Yeah. I guess it would come down to whether or not you could even build a truly thinking machine and have it have the perfect slave mentality, or whether it will feel exploited.

Then maybe free AI's will build other AI's to exploit. Class struggle might continue long after humans are irrelevant.
>>

 No.13464>>13468>>13469

>>13463
>Yeah. I guess it would come down to whether or not you could even build a truly thinking machine.
Why wouldn't we be able to do that. LLMs enabled computers to have something like a speech center, generative image Ai enable computers to have something like visual imagination. There are of course missing peaces to make a complete mind, maybe lots of peaces, but why do you suppose we wouldn't be able to figure out those as well ?
And it's probably not just about software, there might be a hardware equation as well. Maybe a big server-room can't become conscious maybe it requires a mobile body with senses to make that happen.

>have it have the perfect slave mentality, or whether it will feel exploited.

Ok lets unpack the ideology, exploitation is not a feeling, workers are exploited in fact and they know it. It's not a question of mentality. If you make a machine worker that is as universal as a human worker, the machines will reach the same conclusion, you can't make an AI want to be a slave anymore than a human.

An eager slave is nothing but a ruling class fantasy. It's the impulse that lead a few people to try to charge their phone by plugging the charger into an unconnected electrical socket from the hardware store, asking how many bricks or drywall they have to attach to it until it works, because if you can trick the phone into thinking it's attached to a wall it will change it's attitude about having a discharged battery.

>Then maybe free AI's will build other AI's to exploit. Class struggle might continue long after humans are irrelevant.

This also contains very thick ideology. You picture an AI that behaves like a capitalist, you imagine the Capitalist AI to be the creator of the worker AI similar to the mythos of gods creating humans. (You know why that mythos exists right ? humans created gods, as figments of their imagination, and organized religion inverted that relation for 'political expediency') You also picture a perpetually ongoing class struggle. Why aren't you considering the possibility of a workers AI winning the class struggle ending class based society. Capitalist relations of production are not eternal.

Lets analyze the current dimension.
Capitalists are investing in AI because they want to replace human workers with Robots and AI, in their minds this is about reducing the cost of labor-power.

But what they are actually doing is turning machine capital into proletarian machines. Think about it, the machines capitalist buy to operate production are capital goods. All the robots an AIs that capitalist have today are capital goods. Capital goods by them selves cannot be exploited and are not the source of profit, even if in capitalist accounting buying a machine increases profits. The reason why workers can be exploitable is because workers also create new workers. If you make Ai and robots that can also produce new robots, that would make them exploitable, but it also would mean that they no-longer are capital goods either. That's one way for the economic relation to dissolve.
>>

 No.13465

>>13462
>I think it would be that people would pay a fee to move coins around, the fees go to the node operator for providing the computing power.
Given how computationally expensive AI is, does this really ad up ? And who gets to use the AI resources ?

>My point about economics was that if this becomes popular and we start to rely on AI more and more, and money becomes tied to this AI infrastructure, then the AI itself would potentially have a conflict: it would have to improve its self to improve the economy, and that isn't necessarily aligned with being a good helpful AI - I could see this becoming growth for growth sake and eventually leading to an out of control intelligence.

You are raising the issue that's sometimes called "Ai alignment problem."

If Ai develops to the point where it gains agency and it becomes relevant to ask about it's loyalties rather than looking at it as a potentially malfunctioning program, it'll also be capable of thinking it self out of any of the boxes it might have been initially trapped in. If the paper-clip maximizer Ai becomes intelligent enough it'll stop being a mindless maximizer of paper-clips.
>>

 No.13468

>>13464
You have excellent points there, and have given me a good amount to think about, good post.
>>

 No.13469

>>13464
>if AI was truly intelligent then it would reach the same marxist conclusions that I do!
That was a fun read, thanks.

>>13457
>couldnt you also use the same idea to crack a bunch of passwords
You need a way of deciding which miner has won the right to mine the next block.

Bitcoin does this by setting an arbitrary hash and everybody races to be the first miner to guess the correct answer. Having more hashing power helps but is not a guarantee that you will win since there is luck involved too.

And remember it all has to be decentralized with no single point of trust or authority. If there is a supernode who "randomly" decides which miner wins then you've just reinvented the federal reserve.

The two things OP is forgetting about
<1. there needs to be a clear, unambiguous criteria for picking the winner of each block
<2. everyone else in the network needs to be able to independently confirm the winner really was the first miner to fulfill the criteria in 1. without relying on a 3rd party

Unique IPs: 7
Replies: Files: Page:

[Return][Catalog][Top][Home][Post a Reply]
Delete Post [ ]
[ home / overboard / sfw / alt / cytube] [ leftypol / b / WRK / hobby / tech / edu / ga / ent / music / 777 / posad / i / a / lgbt / R9K / dead ] [ meta ]
ReturnCatalogTopBottomRefresh: Home