Getting started with Code Interpreter

If you only read one thing about #AI or #LLMs this week, make it this: What AI Can Do With A Toolbox... Getting started with Code Interpreter by Ethan Mollick

TL;DR:

- OpenAI has finally released Code Interpreter to all #ChatGPT Plus subscribers.
- Beyond visualizations, ChatGPT can now cut down on hallucinations and math/logic errors by writing and running code to generate output, instead of predicting tokens.

Here’s Dan Mason's hot take:

- ChatGPT no longer waits for you at each step with Code Interpreter. Tell it what you want, it brainstorms ideas, and then it will go ahead and write code for the first two on its own! It requires a real rethink of how we do our jobs, and how we measure our own value in the process.
- Code Interpreter has significant gaps — it can’t access the Internet and pushes back when you ask it to infer data you didn’t provide. Also if it doesn’t have a given library it needs to execute its code, it bails out and tells you to do it yourself.