Talking About Large Language Models

Talking About Large Language Models
arxiv.org

Don't think of an LLM as a full "computer" or "brain". Think of it like a CPU. Your CPU can't run whole programs, it runs single instructions. The rest of the computer built around the CPU gives it the ability to run programs.

Think of the LLM like a neural CPU whose instructions are relatively simple English commands. Wrap the LLM in a script that executes commands in a recursive fashion.

Yes, you can get the LLM to do complicated things in a single pass, this is a testament to the sheer size and massive training set of GPT3 and its ilk. But even with GPT3 you will have more success with wrapper programs structured like:

    premise = gpt3("write an award winning movie premise)
    loop 5 times:
        critique = gpt3("write a critique of the premise", premise)
        premise = gpt3("rewrite the premise taking into account the critique", premise, critique)
    print(premise)

You can somewhat emulate a few layers of this without wrapper code by giving it a sequence of commands, like "Write a movie premise, then write a critique of the movie premise, then rewrite the premise taking into account the critique".

The model is just trained to take in some text and predict the next word (token, really, but same idea). Its training data is a copy of a large swath of the internet. When humans write, they have the advantage of thinking in a recursive fashion offline, then writing. They often edit and rewrite before posting. GPT's training process can't see any of this out-of-text process.

This is why it's not great at logical reasoning problems without careful prompting. Humans tend to write text in the format "". So GPT, being trained on human writing, is trained to emit a conclusion first. But humans don't think this way, they just write this way. But GPT doesn't have the advantage of offline thinking. So it often will state bullshit conclusions first, and then conjure up supporting arguments for it.

GPT's output is like if you ask a human to start writing without the ability to press the backspace key. It doesn't even have a cognitive idea that such a process exists due to its architecture and training.

To extract best results, you have to bolt on this "recursive thinking process" manually. For simple problems, you can do this without a wrapper script with just careful prompting. I.e. for math/logic problems, tell it solve the problem and show its work along the way. It will do better since this forces it to "think through" the problem rather than just stating a conclusion first.

I've found it so incredibly useful to simply replace Google."

Heard of Stack Exchange?

I teach and I expect many students to use language models like ChatGPT to do their homework, which involves writing code. Lots of what people are doing with it is coding (there have been quite a few posts here using it that way).

I've actually also used ChatGPT for literary/song writing experiments and it stinks, aesthetically. The lyrics it wrote, even with a lot of prompting, were totally asinine. And how could they not be?

ChatGPT, without any major changes, is already the best tool out there for answering programming questions. Nothing else comes close. I can ask it to provide code for combining two APIs and it will give useful and clean output. No need to trudge through documentation, SEO-hacked articles, or 10 different Stack Overflow answers. Output quality will only improve from here. Does it sometimes make mistakes? Yes. There are also mistakes in many of the top SO answers, especially as your questions become more obscure.

Aside from programming, how many other fields are there where LLMs will become an indispensable tool? I have a PhD and ChatGPT can write a more coherent paragraph on my thesis topic than most people in my field. It does this in seconds. If you give a human enough time, they will be able to do better than ChatGPT. The problem is, we're already producing more science within niche scientific fields than most scientists could ever read. As an information summary tool, I think LLMs will be revolutionary. LLMs can help individuals leverage knowledge in a way that's impossible today and has been impossible for the last 30 years since the explosion in the number of scientific publications.