@tjrobertson52

I wanna talk about LLM psychosis. If you or someone close to you has not already been affected by this, I think you probably will soon. The word psychosis might be a little too strong given how common this is, but apparently that's the term we've agreed on.

If you don't know, this is just when someone becomes overly confident in an idea after having a conversation with a large language model. In extreme cases, people become outright delusional and I believe it's the result of not understanding what these models are actually capable of. So let's talk about what this typically looks like, how to spot LLM psychosis and how to avoid it.

Most common version of this I've seen is a friend or colleague coming to me with their amazing new business idea. It's always a very unique idea. They're always very excited about it and very confident that it's gonna work out.

Most cases of LLM psychosis start with the person bringing their own idea to the large language model. A large language model is then affirming their idea and crucially providing a long list of very smart sounding reasons their idea is definitely great. Without a deep understanding of what these models are and aren't capable of, it's very easy to be taken in by these list of reasons.

And to make matters worse, the LOLM will then typically encourage the user, encourage them to go deeper, to think bigger, to go more outside. The box and this can all be very exciting and very compelling. Most of us crave this kind of validation and human reinforcement has turned these LLMs into validation machines and they're only getting better.

To be able to spot and avoid LLM psychosis, we need to understand what the models are and aren't capable of. If you're an expert in a given field, large language models are really good at giving you ideas. But if you've ever had a conversation with one of these large language models on a topic that you're an expert on, you will realise very quickly that it still often gets things wrong.

And you really need to be an expert to notice this because these models are really good at saying things that sound true. And if you're a complete novice in a given area, these large language models can quickly turn you into an amateur. As long as the stakes aren't too high if you get a few things wrong, what else can't do is let you leap frog past experts.

So if you find yourself disagreeing with experts after having a conversation with a large language model, you might be under LLM psychosis. If you find yourself reading ChatGPT's diagnosis of your symptoms to your physician, you might be under LLM Psychosis. If as a few of my clients have done, you find yourself giving ChatGPT's SEO advice to your SEO agency, you might be under LLM Psychosis.

Of course, these LLs are still very helpful and powerful tools, but the more powerful they become, the more we're gonna have to watch out for LLM Psychosis.