4 ms·

> A ton of thinking and experimentation ... is both breaking things down into small steps, and something that an LLM can help with. > They seem quite impossib

> A ton of thinking and experimentation ... is both breaking things down into small steps, and something that an LLM can help with. > They seem quite impossible to solve until you try a bunch of things, sometimes for years/decades, throwing stuff at the wall Do you see what you did there? That's breaking things down and trying lots of small things. What you seem to think I'm suggesting - which I'm not - is that solving hard problems is linear once it's fragmented. I'm suggesting hard problems are only solvable through fragmentation - breaking them apart - but I'm not in any way suggesting that this means they're solvable through a simple linear thinking process. Fragmentaiton and linearity are not the same thing. If anything, LLMs can help make non-linear thinking more efficient by getting you out of small areas of focus, and as I said originally helping you explore a problem through metaphor, different perspectives, different domains, and so on.

crazygringo 3y ago
> > They seem quite impossible to solve until you try a bunch of things, sometimes for years/decades, throwing stuff at the wall > Do you see what you did there? That's breaking things down and trying lots of small things. No it's not, that's my whole point. If you're trying to find a filament that will work for a commercial light bulb, then testing out 1,000 materials is not breaking anything down to solve the problem. Instead, it's trial and error. They're literally the opposite approaches. Some problems are solvable through fragmentation. Many others are just fundamentally not, and these obviously tend to be the more difficult ones. > If anything, LLMs can help make non-linear thinking more efficient by getting you out of small areas of focus That's literally the opposite of breaking things down into small steps. So now I don't even know what you're arguing anymore. But also, I don't see how LLM's help that in the least. I have not seen any examples of LLM's demonstrating "non-linear thinking". They literally work, well, with linear token prediction -- one token at a time.
- PaulRobinson 3y ago
  Perhaps you could provide an example of a hard problem that was solved that wasn't done so through breaking a problem down into small pieces. No academic paper ever written, invention ever created or piece of art I can think of or heard of (including general relativity, which was referenced earlier in thread), was solved by just sitting there and thinking about the big hard problem and solving the big hard problem in one go. The closest I can think of what you're referring to is an accidental discovery, so I think that's where our lines are crossed. At its core I still don't buy that LLMs are useless in the context of solving hard problems. You disagree. Time and experience will tell, but it seems more likely that successes will be attributed to them than not.

crazygringo 3y ago

> > They seem quite impossible to solve until you try a bunch of things, sometimes for years/decades, throwing stuff at the wall > Do you see what you did there? That's breaking things down and trying lots of small things. No it's not, that's my whole point. If you're trying to find a filament that will work for a commercial light bulb, then testing out 1,000 materials is not breaking anything down to solve the problem. Instead, it's trial and error. They're literally the opposite approaches. Some problems are solvable through fragmentation. Many others are just fundamentally not, and these obviously tend to be the more difficult ones. > If anything, LLMs can help make non-linear thinking more efficient by getting you out of small areas of focus That's literally the opposite of breaking things down into small steps. So now I don't even know what you're arguing anymore. But also, I don't see how LLM's help that in the least. I have not seen any examples of LLM's demonstrating "non-linear thinking". They literally work, well, with linear token prediction -- one token at a time.

PaulRobinson 3y ago
Perhaps you could provide an example of a hard problem that was solved that wasn't done so through breaking a problem down into small pieces. No academic paper ever written, invention ever created or piece of art I can think of or heard of (including general relativity, which was referenced earlier in thread), was solved by just sitting there and thinking about the big hard problem and solving the big hard problem in one go. The closest I can think of what you're referring to is an accidental discovery, so I think that's where our lines are crossed. At its core I still don't buy that LLMs are useless in the context of solving hard problems. You disagree. Time and experience will tell, but it seems more likely that successes will be attributed to them than not.