mcfry

Great comment! I've been trying to understand it more and was hoping to find more people discussing the result, or the implications of the result, and this was the missing piece for me after watching a few videos.

This is cool, but I feel like the information needs to be refined by someone who understands the domain. A long-form educational resource certainly requires many decisions to be made - where to add more/less detail, where to refine, where to better visualize, etc - and humans are still king of the decision domain, for now at least. Even with the visuals, it feels a lot like the step-by-step I might receive from ChatGPT. There's some value add, I guess, because we're visual creatures first and foremost... but it's tentative.

I don't see nearly enough discussion on the distinction between LLM and human reasoning, where the boundary lies, that type of thing. It's like we've all collectively buried our heads in the sand and accepted that they'll eventually be able to think exactly like us.

It's still using a fork of vscode and literally everyone is aware that these other IDEs are forked.

How hard is it to have a video player with a fucking volume toggle?

But it can't actually deduce, can it? If 136891438 * 1294538 isn't in the training data, it won't be able to give you a valid answer using the model itself. There's no process. It has to offload that task to a tool, which will then calculate and return.

Further, any offloading needs to be manually defined at some point. You could maybe give it a way to define its own tools, but even then they would still be defined by what has come before.

A proof assistant is a verifier, and a tool so therefor a patch, so I really fail to see how that could be understood as the LLM having deduction.

Something which I haven't been able to fully parse that perhaps someone has better insight into: aren't transformers inherently only capable of inductive reasoning? In order to actually progress to AGI, which is being promised at least as an eventuality, don't models have to be capable of deduction? Wouldn't that mean fundamentally changing the pipeline in some way? And no, tools are not deduction. They are useful patches for the lack of deduction.

Models need to move beyond the domain of parsing existing information into existing ideas.

This is just... rebranding for instructions and files? lol. Love how instructions for creating a skill is buried. Marketing go brr.

You still have the 'pointy' problem, even with many layers, no? The bottom-most block has to be a triangle.

Was thinking the same thing. It's impossible to take this article's criticisms of AI seriously when it's so obviously over-edited with AI itself.