How to make robots think good

GM, this is Neural Nonsense. Giving you the top trends in AI so you can automate your job before your boss does.

Here's what we've got for you today:

• Making robots think good

• Launch your own AI endpoint in seconds

• Get virtual office hours with a tutor on any subject

🤖 How to make robots think good

If you've played around with ChatGPT for long enough, something becomes clear. It can be wrong, and you know what's worse?

It can lie.

Now lying, apart from being noticeably absent from Azamoth's laws of robotics, is also kind of a bad thing. We want good robots.

But this is actually a really hard problem. When you ask an AI to explain an answer, what's actually happening is the AI is making up a story about that answer. That's how you end up with Dinosaur civilisations.

Until now.

A new paper has hit the mean streets of the internet with a new way of writing prompts, called "Self Ask". How it works is to ask the model to break a question into smaller bits, and answer those questions first.

Here's a handy video explaining it in more detail.

🛠️ Tools and Tidbits

• Maggie Appleton wrote a compelling piece about the coming "dark forest" and how to signal you are human in a sea of ghosts [link]

• Scale AI - You can now create and launch a hosted AI API in seconds with Scale AI [link]

• Riff - Get virtual office hours for any subject created by a single student at university [link]

🎨 DALL-E of the Day

The iPhone in 2023 by saint laurent del rey [Link]

So, how did I do?

This is my first ever newsletter, so I'm sorry if it wasn't up to snuff! Let me know exactly what I could do better by replying this email, or messaging me on Twitter.