Large language models

This page has 0 threads | Add post

Blogs

Finding our blogs useful? Please consider helping us to share more!

What you can - and can't - hope to understand about how AI works Part three of a four-part series of blogs
I'm sick of reading blogs pretending to explain how Large Language Models work, when the truth is that few people on earth can ever hope to understand how they are constructed. This blog is my attempt to explain as much as 99.99% of us can ever hope to understand about how AI works, and why the other bits will always remain opaque to most people. What you can - and can't - hope to understand about how AI works Tokenisaton and embedding Large language models (this blog) Constraints on the use of AI

What you can - and can't - hope to understand about how AI works

Part three of a four-part series of blogs

I'm sick of reading blogs pretending to explain how Large Language Models work, when the truth is that few people on earth can ever hope to understand how they are constructed. This blog is my attempt to explain as much as 99.99% of us can ever hope to understand about how AI works, and why the other bits will always remain opaque to most people.

What you can - and can't - hope to understand about how AI works
Tokenisaton and embedding
Large language models (this blog)
Constraints on the use of AI

Posted by Andy Brown on 16 July 2024

In this blog

Building a Large Language Model
Transformers and Attention
Why am I being so negative?

So this is where our journey of understanding ends - well, almost.

Building a Large Language Model

Here's a simplistic (and believe me, simplicity is good when consdering AI tools) view of the steps involved in building a large language model:

Step	Description
1	Assemble lots of data (the more the better). One of the most obvious potential sources of data might be the text on the world's websites, although this raises questions of bias and of ownership (as discussed in the last part of this blog).
2	Parse it to build a large language model. This involves not just building up a list of all the tokens used (see previous part of this blog), but also creating a list of parameters or dimensions and where each token belongs within that parameter's space. Parameters/dimensions might include things like how important a word is or where it might appear in a sentence, but in truth only the model itself is likely to understand all of the parameters created.
3	Train this model (for example by asking it questions and grading its responses to get it to fine-tune its parameters and data).

The process of building and training a large language model is famously expensive, requiring huge inputs of data, computer processing power and electricity.

So how does an organisation like OpenAI, Gemni or Antropic build a large language model?

Transformers and Attention

Transformers provide the T in GPT (the acronym stands for Generative Pre-trained Transformers). Invented by Google as recently as 2017, a transformer involves recalibrating the input embedded vectors (remember them?) representing your question by applying the principle that some words in a sentence are more important than others. If you think that sounds straightforward, here's a core diagram from the Wikipedia page on attention in machine learning:

You're going to have to be good at vector spaces, tensors, probability distributions, vectors and matrices to understand this (and those are just the words I recognised in the articles on this subject). Most of these are things you would only cover properly as part of a maths or physics degree.

There are a lot of blogs trying to explain how large language models work, but they nearly all have the hallmark of having been created by AI tools, or just copying what is written elsewhere. I suspect that the number of people in the world who truly understand how large language models work could be numbered in the tens of thousands, or even just thousands.

Why am I being so negative?

If it sounds like I'm trying to put you off learning about this, that's because ... I am. I have no idea how many things work in life, beginning with our cat, but in the modern world you just have to accept that you're never going to understand many things and move on. I don't have the time or intellectual capacity to understand the maths behind the Transformer algorithm, and I would make a bet that you don't either!

Parts of this blog
What you can - and can't - hope to understand about how AI works Tokenisaton and embedding Large language models (this blog) Constraints on the use of AI

Some other pages relevant to the above blogs include:

AI (Artificial Intelligence) classes

This blog has 0 threads

Add a new post

BLOGS BY TOPIC

BLOGS BY AUTHOR

Building a Large Language Model

Transformers and Attention

Why am I being so negative?

Head office

London

Manchester