Simon Willison’s Weblog

The Changelog podcast: LLMs break the internet

I’m the guest on the latest episode of The Changelog podcast: LLMs break the internet. It’s a follow-up to the episode we recorded six months ago about Stable Diffusion.

This time, we spent the whole episode talking about large language models: ChatGPT, GPT-4, Bing, Bard, Claude, LLaMA and more.

I listened to this again today while walking the dog. It’s good! It’s the best representation of my current thinking about this wild AI-enhanced world we are rapidly entering.

We start the episode by reviewing my predictions from six months ago. I said that search engines like Google would have LLM features within two years—Bing and Bard are live already, so I over-shot on that one. I also said that there would be LLM tools for creating 3D worlds within six months. When we recorded the episode last week I hadn’t seen any that quite matched my prediction... and then yesterday Pete Huang posted a Twitter thread listing six of them!

There’s a lot of other stuff in there: the full episode is 1 hour and 40 minutes long.

I’ll quote one section in particular, from part way through my answer to the question Where should someone start with this? (direct link to audio).

This is the thing I worry that people are sleeping on. People who think “these language models lie to you all the time” (which they do) and “they will produce buggy code with security holes”—every single complaint about these things is true, and yet, despite all of that, the productivity benefits you get if you lean into them and say OK, how do I work with something that’s completely unreliable, that invents things, that comes up with APIs that don’t exist… how do I use that to enhance my workflow anyway?

And the answer is that you can get enormous leaps ahead in productivity and in the ambition of the kinds of projects that you take on, if you can accept both things are true at once at once: it can be flawed, and lying, and have all of these problems… and it can also be a massive productivity boost.

Here are four illustrative examples of things I’ve used LLMs for as a huge productivity booster in just the past few weeks.

I also gave my review of Google Bard at 1:14:46 which I think deserves a listen.