It can be helpful for quickly summarizing a vast body of knowledge or a highly complex topic, to get a general overview and see which strings to pull further, as long as you don’t take everything at face value and understand that you still need to pull those strings yourself in order to acquire an understanding.
Like, if I suddenly wanted to learn computer programming, I wouldn’t know where to start. But querying an LLM can give me a general idea, define a few key terms and explain the difference between related concepts, without me having to browse through a hundred different tech blogs to answer all my questions in terms I can understand.
But I wouldn’t suddenly think I’m a computer programmer after doing that. I would have a better idea of where to start learning. I would be able to decide whether to focus first on object-oriented programming or functional programming, static or dynamic typing, declarative or imperative syntax, etc., instead of getting overwhelmed from the start just trying to learn the differences between those concepts.
It can also suggest resources for further learning, books or websites written by humans, links to open-source software that does what I’m trying to do, etc.
I wouldn’t expect it to write code for me, but it can be an efficient aid to self-learning and show me what programs and libraries to use for my intended purpose.
Or for astrophysics, for example. I wouldn’t expect it to give me an accurate breakdown of the engineering specs required to build a pair of O’Reilly cylinders at a Lagrange point, but it can suggest software for rendering prototypes or for simulating the forces that need to be accounted for.
That wouldn’t make me an astrophysicist, but it’s kind of cool that you don’t need to be one to learn about this stuff and tinker around in a field that’s so vast and technical as to be otherwise prohibitive for non-experts.
It also depends on the LLM of course. I think Mistral and Lumo are generally pretty okay at doing what I described above. Their algorithms aren’t corrupted by american venture capital, at least, so they have more incentive to give you an accurate response rather than being sycophantic and hugboxing.
It can be helpful for quickly summarizing a vast body of knowledge or a highly complex topic, to get a general overview and see which strings to pull further, as long as you don’t take everything at face value and understand that you still need to pull those strings yourself in order to acquire an understanding.
Like, if I suddenly wanted to learn computer programming, I wouldn’t know where to start. But querying an LLM can give me a general idea, define a few key terms and explain the difference between related concepts, without me having to browse through a hundred different tech blogs to answer all my questions in terms I can understand.
But I wouldn’t suddenly think I’m a computer programmer after doing that. I would have a better idea of where to start learning. I would be able to decide whether to focus first on object-oriented programming or functional programming, static or dynamic typing, declarative or imperative syntax, etc., instead of getting overwhelmed from the start just trying to learn the differences between those concepts.
It can also suggest resources for further learning, books or websites written by humans, links to open-source software that does what I’m trying to do, etc.
I wouldn’t expect it to write code for me, but it can be an efficient aid to self-learning and show me what programs and libraries to use for my intended purpose.
Or for astrophysics, for example. I wouldn’t expect it to give me an accurate breakdown of the engineering specs required to build a pair of O’Reilly cylinders at a Lagrange point, but it can suggest software for rendering prototypes or for simulating the forces that need to be accounted for.
That wouldn’t make me an astrophysicist, but it’s kind of cool that you don’t need to be one to learn about this stuff and tinker around in a field that’s so vast and technical as to be otherwise prohibitive for non-experts.
It also depends on the LLM of course. I think Mistral and Lumo are generally pretty okay at doing what I described above. Their algorithms aren’t corrupted by american venture capital, at least, so they have more incentive to give you an accurate response rather than being sycophantic and hugboxing.