2 ms·

It's a process, and being contemptuous about each small step isn't helpful; that attitude doesn't help us find "the trick". LLMs represent an incredible leap i

by ta8645 3y ago

It's a process, and being contemptuous about each small step isn't helpful; that attitude doesn't help us find "the trick". LLMs represent an incredible leap in abilities, and offer untold opportunities to provide services that no other technology we currently possess can offer. Some people definitely overstate our current progress, but it seems a lot more just want to shit on it, and undervalue it instead. And what exactly did you mean by: "Computer Science still puts the same restrictions on what we can compute." Because it sure sounded like you were saying that there is an absolute limit on our ability to produce intelligent systems. Which is why I started talking about human abilities, as an instance proof of the capabilities we should be able to offer.

mgaunard 3y ago
"An incredible leap", more like an over-hyped trend to get money out of venture capitalists. In practice it's a recipe for spending large amounts of money on GPU cloud compute to achieve things that could have been done just fine with more traditional methods. > And what exactly did you mean by: > "Computer Science still puts the same restrictions on what we can compute." Computer Science is the science of studying what can be computed, what is the complexity of those algorithms, what are the limits of our logic and mathematical models. We know there are some true statements that cannot be proved to be true. We know there are algorithms that we cannot prove whether they terminate. We know the complexity bounds in solving a number of problems or getting suitably precise approximations. Our computers are bound by the same constraints than a Turing machine is, and those constraints are well-studied. LLMs haven't changed the capabilities of our computing systems.
- ta8645 3y ago
  > In practice it's a recipe for spending large amounts of money on GPU cloud compute to achieve things that could have been done just fine with more traditional methods. There are no traditional methods that as convincingly write poetry, sing songs, animate movies, or synthesize search results. You should be able to lament the hype, while still acknowledging this truth. > Computer Science is the science of studying what can be computed, what is the complexity of those algorithms, what are the limits of our logic and mathematical models. Computer science is too limited to properly decide the matter. We know for a FACT that physics allows human level cognition (obviously, since we exist). Therefore, all that is required is for us to understand and harness those rules of physics, and we can manufacture devices that match or exceed human level cognition. Given enough time, there is no logical reason we can't achieve this result. People who continue to insist that it is logically impossible, are short-sighted, and blinded by human vanity.

by ta8645 3y ago

mgaunard 3y ago

"An incredible leap", more like an over-hyped trend to get money out of venture capitalists. In practice it's a recipe for spending large amounts of money on GPU cloud compute to achieve things that could have been done just fine with more traditional methods. > And what exactly did you mean by: > "Computer Science still puts the same restrictions on what we can compute." Computer Science is the science of studying what can be computed, what is the complexity of those algorithms, what are the limits of our logic and mathematical models. We know there are some true statements that cannot be proved to be true. We know there are algorithms that we cannot prove whether they terminate. We know the complexity bounds in solving a number of problems or getting suitably precise approximations. Our computers are bound by the same constraints than a Turing machine is, and those constraints are well-studied. LLMs haven't changed the capabilities of our computing systems.

ta8645 3y ago
> In practice it's a recipe for spending large amounts of money on GPU cloud compute to achieve things that could have been done just fine with more traditional methods. There are no traditional methods that as convincingly write poetry, sing songs, animate movies, or synthesize search results. You should be able to lament the hype, while still acknowledging this truth. > Computer Science is the science of studying what can be computed, what is the complexity of those algorithms, what are the limits of our logic and mathematical models. Computer science is too limited to properly decide the matter. We know for a FACT that physics allows human level cognition (obviously, since we exist). Therefore, all that is required is for us to understand and harness those rules of physics, and we can manufacture devices that match or exceed human level cognition. Given enough time, there is no logical reason we can't achieve this result. People who continue to insist that it is logically impossible, are short-sighted, and blinded by human vanity.