AI Horseless Carriages

Summary

Summarized wtih ChatGPT

The author argues that AI apps, like Gmail’s email assistant, often produce awkward results because they don’t let users customize the system prompts. Users should be able to write and edit their own system prompts to improve the AI’s performance and better reflect their personal style. The goal of AI should be to automate mundane tasks, making users' lives easier.

Key Takeaways:

1. Allow users to create and edit their own system prompts for better AI output.
2. Focus on designing AI that automates tedious tasks instead of just mimicking human work.
3. Encourage developers to provide user-friendly interfaces for prompt writing and feedback.

Highlights from Article

Combined Prompt System Prompt: Instructions that define the assistant’s behavior… User Prompt: Specific query or instruction from the user.

• definition of a system and user prompt

Of course I’m being glib here, but the problem is not just that the Gmail team wrote a bad system prompt. The problem is that I’m not allowed to change it.

• Some of the system prompts take over too much control, like the Gmail email writing

There’s something magical about teaching an LLM to solve a problem the same way you would and watching it succeed. Surprisingly, it’s actually easier than teaching a human because, unlike a human, an LLM will give you instantaneous, honest feedback about whether your explanation was good enough or not. If you get an email draft you’re happy with, your explanation was sufficient. If you don’t, it wasn’t.

• Good AI is fun to train

The “old world thinking” that gave us the original horseless carriage was swapping a horse out for an engine without redesigning the vehicle to handle higher speeds. What is the old world thinking constraining these AI apps?

• Old thinking is still pervasive in modern apps

With this framing, it’s only natural to assume that it’s the developer’s job to write the System Prompt and the user’s job to write the User Prompt. That’s how we’ve always built software. But in Gmail’s case, this AI assistant is supposed to represent me. These are my emails and I want them written in my voice, not the one-size-fits-all voice designed by a committee of Google product managers and lawyers.

The division of labor is clear: developers decide how software behaves in the general case, and users provide input that determines how it behaves in the specific case.

• the reason for old world behavior si that developers assume that the system prompt is the backend, so they control it.

All material owns to the authors, of course. If I’m highlighting or writing notes on this, I mostly likely recommend reading the original article, of course.

See other recent things I’ve read here.