Simon Willison.net: Think of Language Models Like ChatGPT as a “Calculator for Words”

I like to think of language models like ChatGPT as a calculator for words.

The more I use these tools, the more I think the narrative around "Microsoft makes Google dance by adding GPT to Bing," narrative is more and more hilarious. Search is so much better than these LLMs…they're more reliable, they're faster, they're easier to parse…everything is better.

But I'm finding tons of use in using Raycast AI to explain things I'm looking at on my screen, using GitHub Copilot to help me write code, or Notion AI to brainstorm video titles inline in a document.

Here We Go Again

Chatbots had a moment back in 2016 as well, with people like Satya Nadella saying, "bots are the new apps," and that basically everything we did from shopping to vacation planning would happen in a chat interface, not a visual UI like we had gotten used to. This is an example of the general vibe as written in the Washington Post:

When you start a conversation with H&Ms bot on Kik, for example, it kicks things off by trying to get a sense of your personal style. It shows you three pairs of outfits, asking you to decide in each set which look you like better. From there, users can go back and forth with the bot asking it to show different outfits centered around a certain type of item

Today, H&M's site is chatbot free because people didn't care for shopping like this, and frankly, it's baffling to me why people seem to be falling in the same trap again. Yes, these new chatbots are technically superior to the ones the same people were fawning over in 2016, but the fact remains that doing things like shopping with text prompts absolutely sucks, no matter how clever the bot responding to you becomes.

Whenever someone says, "these graphical user interfaces are old hat, the future is all about really smart terminal prompts," you should run the other direction because that's a hilarious take.