---

My position on generative AI

tl;dr I think that the generative AI industry is disgustingly amoral, damaging to society and the environment, and should be left to rot.

What is "generative AI"?

For the purpose of this document, I am referring to technology deployed by companies such as Google, Microsoft, Anthropic, OpenAI, X, etc. which is based on large language models (LLMs). Such technology is often used to gaslight unsuspecting people, as well as to waste the time of well meaning software engineers. Much of my criticism can be equally levelled at the technologies which do similar with images, video, and sound.

Large language models are, in very simple terms, the distillation of all the input text given to train them into a huge matrix of numbers which represent all the relationships between snippets of that text (typically words, or part words). They are therefore a derived work of all the text which went into them which raises a number of concerns around the use of their output.

Querying these models requires producing a textual prompt which is reduced to numbers representing those same snippets, and then processing that through the model, producing a prediction of which snippet would come next. It is to all intents and purposes a text autocomplete engine. Since that is entirely stable (same input produces same output) LLMs also include a "heat" value which is purely an amount of randomness added to the output. This means that the predicted next token will have an amount of pure randomness applied to it as well.

As a result there is no reasonable way for a human to predict what the output of these models will be. There are clearly emergent properties of the system which are beyond basic analysis and the inherent dependency on randomness introduces a further inability to trust what such a tool produces.