Post B6Ye80L4HKyw77ofOC by sickburnbro@poa.st
(DIR) More posts by sickburnbro@poa.st
(DIR) Post #B6YV4BUZyAQlxZMugK by sickburnbro@poa.st
1 likes, 1 repeats
another sign. I think this AI stuff is going to run into hard reality soon which will be that token prices are going to have to 10x or 100x which will destroy a ton of use cases.
(DIR) Post #B6Ybqszc8cJmV1UUme by petra@poa.st
1 likes, 0 repeats
Thing is AI has managed to prove its value. Prove what it can accomplish and increase efficiency. That's not going away. It's just its current implementation is so inefficient. This is an engineering problem. A human brain runs on 20 watts and weighs about 3 lb. So we know absolutely it's possible to have human level AI that operates at such minimal resources. We just have to figure out how to do it. The first one to do so takes all the marbles.
(DIR) Post #B6YbqtAxSRON4CnYSu by sickburnbro@poa.st
0 likes, 0 repeats
I mean this is like saying "computers have proven their value" .. which is duh.
(DIR) Post #B6YbwOhLgYuyHPZjnM by sickburnbro@poa.st
1 likes, 0 repeats
the upside and downside of generators is that they are stateless, so every time you do something like asking is a question about any document, that entire document has to be fed in.
(DIR) Post #B6YctqUM0MgDYVqPHk by petra@poa.st
0 likes, 0 repeats
A machine with human level intelligence and agency is quite a bit more than a "computer".It's like saying, "Keyboards have proven their value," referring to typewriters, when microcomputers first made their appearance.
(DIR) Post #B6YctqeHPScU3IUKky by sickburnbro@poa.st
0 likes, 0 repeats
but current LLMS have neither.
(DIR) Post #B6Ye808J2mm1TXqTUu by petra@poa.st
0 likes, 0 repeats
They do.
(DIR) Post #B6Ye80L4HKyw77ofOC by sickburnbro@poa.st
0 likes, 0 repeats
I think you have a mistake in how you are defining agency to begin with. But most critically, intelligence isn't simply facts but a system to align facts in a consistent structure. If you were to model human intelligence as a LLM, it would resist adding in information which broke that structure, but it doesn't.It's simply a next token generator without an ability to query its internal state. You can verify this because if it was, it wouldn't "lose" the ability to do math.
(DIR) Post #B6YrCb9OecTYqVsU7M by BroDrillard@nicecrew.digital
0 likes, 0 repeats
I've been saying for some time that once AI is properly priced in some (many?) roles humans may end up being cheaper.
(DIR) Post #B6YrCbIy528FKCM82K by deprecated_ii@poa.st
0 likes, 0 repeats
@BroDrillard @sickburnbro the likely outcome is all this AI shit gets siloed away in a few megacorps and of course governments because nobody else can afford it, and the large scale infrastructure never materializes because there's no demand at 100x the current pricethat might be scary if it wasn't self-defeating garbage that's guaranteed to hollow out an organization in 5-10 years
(DIR) Post #B6ZDtOKUOdXaWesQDY by AuntNorma@poa.st
1 likes, 0 repeats
It can’t be done with silicon chips. Like you can’t make an airplane out of lead. Only organic carbon is capable of that kind of flexibility and connectiveness.
(DIR) Post #B6ZDtZ4aS6M9qWWNSS by petra@poa.st
1 likes, 0 repeats
Or quantum based processing. There is no reason AI-like performance can't be accomplished with different architectures. Perhaps something completely different from the neural interconnection model we've based the existing machines on. That is very exciting to consider. Skip the neuron model entirely.
(DIR) Post #B6ZDuCswQooBJzNdrM by GenesRus@poa.st
1 likes, 1 repeats
Then why not ask ChatGPT to build itself out of organics and get conscious? And please don’t eat me!
(DIR) Post #B6ZDwWLb2hhKGGt2sS by BroDrillard@nicecrew.digital
1 likes, 1 repeats
LLMs are trained on the entire internet and books and shiet. An individual human brain contains only a tiny fraction of all that.An LLM with a smaller, more targeted knowledge base would be cheaper computationally. But most of the cool capabilities of LLMs emerge when they're yuge. 🤷♂️
(DIR) Post #B6ZDyP7St2ugQKAuau by AGARTHA_NOBLE@shortstacksran.ch
0 likes, 0 repeats
@sickburnbro ...No, that's not right.
(DIR) Post #B6ZDyPI6FVQ6xJ9PAe by sapphire@shortstacksran.ch
1 likes, 0 repeats
@AGARTHA_NOBLE @sickburnbro my guy at work with a PhD in it said 10x is optimistic when I talked to him so I’m gonna say it is
(DIR) Post #B6ZDyZ0k6tTh6N6gVM by sapphire@shortstacksran.ch
1 likes, 0 repeats
@petra @sickburnbro you’re not gonna beat 4 billion years of evolution in a human lifetime, especially when the energy cost outstrips the amount of accessible energy we have left with the tech we’re willing to work with
(DIR) Post #B6ZDz0lSvHtXB6Asa0 by AGARTHA_NOBLE@shortstacksran.ch
1 likes, 0 repeats
@sapphire @sickburnbro With traditional transformer models at 1T or 2.5T, yeah, maybe, but it sounds kinda like your guy is doomering just like everyone else, doesn't know what he's talking about. Never underestimate the world's ability to get better, even at the expense of your expectations.
(DIR) Post #B6ZDzPbOZrh6D6egr2 by sapphire@shortstacksran.ch
1 likes, 0 repeats
@AGARTHA_NOBLE @sickburnbro the guy has a ten year degree in this shit and is incredibly invested in it working, he’s just not huffing copium like you are
(DIR) Post #B6ZDzgcNJTZwz4FDF2 by AGARTHA_NOBLE@shortstacksran.ch
1 likes, 0 repeats
@sapphire @sickburnbro >Look at my quant. You notice something? That's right he's got a PIECE of PAPER in TENSORFLOW
(DIR) Post #B6ZE00c2yPCSzqFEVU by AGARTHA_NOBLE@shortstacksran.ch
1 likes, 0 repeats
@sickburnbro @petra Right on function, wrong on architecture.