101–110 of 412 posts
> It gives an answer with complete confidence, and I sort of believe it. And half the time, it’s completely wrong. That's bullshit, unless you are asking questions specifically designed to make GPT-4 hallucinate. For most real-world, everyday topics, the accuracy is close to 100%. GPT-4 would be utterly useless otherwise.
Earlier quoted context omitted.
Why not? Your brain is already making predictions about what you expect to see and hear as a part of your perception of reality anyways.
Right, but LLMs suggest that learning how to predict and only training on predictions is sufficient to learn anything and to have emergent generative abilities. What if learning to predict the upcoming input is all that's all that is needed for general human intelligence? What if it is all that any animals do?
I read the book Kingdom of Speech a few years ago and that also left me with the perspective that perhaps language has a lot more to do with how we think and perceive the world than most people like to admit. The book has been heavily criticized but I believe it made an interesting point about language.
This is a perfect case of perfection being the enemy of the good.
Useful AI is here. Hard stop.
The impacts will be huge and unpredictable.
Billions will be made.
The world will change.
Humans will continue making rapid progress, via merging various AI methods and new breakthroughs.
Nevertheless, I enjoy reading the debate.
But anyone wringing their hands over how much is GPT thinking is missing the point.
These are companies making products.
This is not academic research.
It's just another tool in a long list of tools made by humans.
And it's already a productive tool.
This reminds me of the skepticism surrounding electric cars while Tesla was already growing by leaps and bounds.
The ship has sailed. The revolution has started. Progress will undoubtedly be rapid and continual.
> The large language models are a little surprising. I’ll give you that. I think this is the key point about LLMs that kind of explains the wide and polarized views on whether it understands or parrots, whether it can think or is the precursor to thinking or is a dead-end, whether it will catastrophically destroy the world, or “merely” make it steadily worse with bullshit, or just put a few industries out of a job. A…
> My current (tentative) resolution of the surprise is that language encoded way more information about reality than we thought it did. (Enough information that you can fully derive reality from language seems improbable, but iirc it did derive Othello and partly derived chess and I would have thought there wasn’t enough information in language to derive those without playing the games as well, so I can’t rule it out…
Earlier quoted context omitted.
> My current (tentative) resolution of the surprise is that language encoded way more information about reality than we thought it did. (Enough information that you can fully derive reality from language seems improbable, but iirc it did derive Othello and partly derived chess and I would have thought there wasn’t enough information in language to derive those without playing the games as well, so I can’t rule it out…
> it's "read" a significant fraction of everything we've ever written about chess I doubt heavily that a significant fraction of chess's writings are even available in digital format, much less inside of CommonCrawl and correctly trained on.
Earlier quoted context omitted.
> language encoded way more information about reality than we thought it did Language is roughly what separates humans from other apes ... so why would it surprise us that it encodes much of the information of civilization?
> so why would it surprise us that it encodes much of the information of civilization? The Humanities are largely considered superfluous in the tech world, why are you surprised that they are surprised?
> The large language models are a little surprising. I’ll give you that. I think this is the key point about LLMs that kind of explains the wide and polarized views on whether it understands or parrots, whether it can think or is the precursor to thinking or is a dead-end, whether it will catastrophically destroy the world, or “merely” make it steadily worse with bullshit, or just put a few industries out of a job. A…
I think you are close to the mark, but you have been subtly mislead: language is not the data we are working with. We are working with text.
Once you fix that particular failure of word choice, everything else becomes much more clear: text contains much more information than language.
We aren't dealing with just any text, either: that would be noise. We're training LLMs on written text.
Natural language is infamous for one specific feature: ambiguity. There are many possible ways to write something, but we can only write one. We must choose: in doing so, we record the choice itself, and all of the entropy that informed it.
That entropy is the secret sauce: the extra data that LLMs are sometimes able to model. We don't see it, because we read language, not text.
The big surprise is that LLMs aren't able to write language: they can only write text. They don't get tripped up reading ambiguity, but they can't avoid writing it, either. Who chooses what an LLM writes? Is it a mystery character who lives in a black box, or a continuation of the entropy that was encoded into the text that LLM was trained on?
Earlier quoted context omitted.
> Making that ocean deeper is not a trivial problem that we can just throw more compute or data at. You can't possibly know that, given that we don't actually understand how LLMs work on a high level. > We've pretty much tapped out that depth with GPT4 GPT-4 is three months old and you're confident that its working principle cannot be extended further? Where do you get that confidence from?
Sam Altman said it himself. He seems like a reasonable source. If you're familiar with other fields of AI, adding more and more layers to ResNet was the hotness for awhile, but the trick stopped working after awhile.
I get why AI people are at pains to say that GPT-* isn’t AI, and that agi is still a long, difficult way off, I do understand that the distinction is important. But ChatGPT has become such a useful tool to help explain a concept or to filter thoughts through I don’t really care if it’s proper AI or just playing pretend. Google search gives me pretty useless results these days, forums are slow and inconsistent to resp…
Absent the goals constantly shifting, GPT-3 can be viewed as one, GPT-4 even more so. You can ask it questions about almost anything (at a broad level) and get an answer. That's what makes it general and "intelligent"
This quote in particular stood out as ignorant:
“What the large language models are good at is saying what an answer should sound like, which is different from what an answer should be.”
That's... not at all how large language models work. Tiny, trivial, toy language models work like this, because they don't have the internal capacity to do anything else. They just don't have enough parameters.
Stephen Wolfram explained it best: After a point, the only way to get better at modelling the statistics of language is to go to the level "above" grammar and start modelling common sense facts about the world. The larger the model, the higher the level of abstraction it can reach to improve its predictions.
His example was this sentence: "The elephant flew to the Moon."
That is a syntactically and grammatically correct sentence. A toy LLM, or older NLP algorithms will mark that as "valid" and happily match it, predict it, or whatever. But elephants don't fly to the Moon, not because the sentence is invalid, but because they can't fly, the Moon has never been visited by any animal, and even humans can't reach it (at the moment). To predict that this sentence is unlikely, the model has to encode all of that knowledge about the world.
Go ask GPT 4 -- not 3.5 -- what it thinks about elephants flying to the moon. Then, and only them go write a snarky IEEE article.