thrw045

Maybe someone can clarify this:

He says:

> A reasoning model is essentially an upgrade from a conventional LLM, that is, adding more post-training via reinforcement learning with verifiable rewards (and optionally inference scaling techniques) to make it more capable of solving more challenging problems.

But this isn't really true is it? When reasoning models came around with OpenAI's o1, the main feature it had was to be able to try several paths and select the best one. This is what a reasoning model is. This was before agents and reinforcement learning with verifiable rewards wasn't it? And then RLVR and agents were added on later. Or am I wrong?

Yeah I understand your point because I've made music and design for 25 years the old "manual way", and there is definitely less friction with AI than there is the manual way.

But what interests me about AI is that it is possible to do things that you couldn't do before particularly in processing images. For these images I took photos of the bottles myself and used other images and the AI can place the camera anywhere in a scene and place objects anywhere in the scene, and things like that. You can create almost any image you can imagine with the right sources and the right prompt, and even here there is some friction because many images do end up looking like slop and you have to prompt it right to get something non-slop like out of it (though not always).

This ability to manipulate and process information is what makes AI interesting to me, although I'm not sure I adequately explained it. It feels like a new tool like Photoshop except across a different dimension. I feel like we've barely scratched the surface of what's possible to create.

That's fair but there is a lot of creativity in prompting and finding the right image because on the way to those images there is a lot of experimentation and I came up with the ideas for the images not the AI.

It's not needed for a professional designer / photographer etc but now a person who does not have those resources can create them

Sure it may be that way now. But if you go back 20 years to for example video game programming, the hard part was figuring out how to solve the problem, not spend weeks or months trying to figure out the abstract design of the code itself. 10-20 years ago a lot of the value of programming was figuring out how to do something, which is by its nature about writing the code, I would argue.

Edit: Also, even outside of video games, I don't remember people being obsessed about software design 10-20 years ago the way they are now. You had some design principles of course and there were obviously bad or redundant ways of doing things, but it's on a whole other level the past 5 years at least from what I remember.

This is a rewriting of history that I don't really like. Back before LLMs and before that again, it was very hard to write code. That's why so many people asked questions and figuring out how to write a piece of code was a big part of the value add of a programmer. It separated the ones who knew their stuff from the ones who didn't. Yeah sure software design was also important but it was less important than the actual writing of the code, which is why a lot of old code had bad designs. It took decades to figure that stuff out.

Sure, we can't tell, but that's a really bad argument to make the claim that it _is_ conscious. I just have a really hard time believing that a transformer can be conscious, but yes I can't prove it.

It depends on what you mean by alignment. If you mean the sense it's used now, meaning bias in models, security and safety for users and so on, then that's a way more mundane version of alignment than what Yudkowsky and ilk promoted. They are talking about a superintelligent being possibly destroying humanity / the earth. And it wasn't that hard to predict. We have many examples of more technological / "advanced" people ruling or extinguishing weaker people. So a superintelligent AI being a threat is not a huge leap.

IMO they have been consistently too aggressive on timelines. When GPT-4 came out, Yudkowsky said it might be conscious. I think he has written interesting stuff but let's be real

You should read the whole thing. At the end he switches the argument on to himself and says that one should always ask questions, put the ego away and try to get better. He already made the point you made.

Personally I dislike all of these new formats. Shame that most big sites use either avif or webp. When I save these images to disk I always convert to png because storage isn't an issue and bandwidth isn't an issue. Oh well...

Depending on how complex the code is, you don't need that big prompt with ChatGPT.

I have sped up a project by simply saying "What are all the possible ways I can speed up this code?" Then it'll list everything it finds, then ask it to rewrite the code.

Edit: Also, I find I didn't need to do this (because a speed up implies semantic similarity), but you can also add "change it without altering the semantics of the code" and in this way it'll be the same and should pass tests

Ah yeah jagged intelligence is the perfect phrase for it. I do also get some value from them, both in coding and in images. I find it the least usable for information primarily because of the hallucination problem. I still do use it for that purpose but it's kind of annoying when it writes something that's wrong, and I find it out from a Google search later.