Is prompt engineering here to stay?

Welcome to another edition of what we’re determined to make the best damn newsletter in AI. Here we’ll break down AI topics that matter, open your mind to use cases, and keep you ahead of the curve.

Our #1 goal is to be useful. So please shoot us an email 📩 if you have questions or feedback, and especially if you implement something we share!

Here's what we're covering today:

• Our advice for people and organizations in prompting

• Metadata is great to have, but cumbersome to create - so use AI instead

• A look at AI and its impact on engineering, real estate, and even education

Let’s get to it! 👇

ANNOUNCEMENT

Membership pre-launch!

We're here to help you adopt AI faster and our next step is creating a high value community for everyone who is serious about applying AI to their work and businesses.

This newsletter will always be free.

But if you're ready to go all in on AI, we're launching this for you. Check out the details here.

TODAY'S PERSPECTIVE

Prompting for your use case should still be top of mind

The prompt engineering space is heating up. Companies like Anthropic, an OpenAI competitor, are posting full-time Prompt Engineer job listings.

And this trend isn’t just limited to full-time employment. Contractor job postings on Upwork related to specialized prompt creation for company-specific use cases continue to be added daily.

Our informed "opinion"? (Yes, still opinion. We are so extremely early in this tech).

Current prompting trends have highlighted a "bag of tricks" approach to prompting, which focuses on obscure or otherwise overly complicated syntax to generate the desired responses.

But AI researchers are continuously working on making AI systems "align" with our goals. ChatGPT is an excellent example of this. As we've covered before, ChatGPT was trained to be good at chat through a process called reinforcement learning from human feedback. If you are using GPT3 (the model behind ChatGPT), you do need to do a little more prompting to get good results!

So, it's very likely that AI models will get better and better at understanding our goals in plain language over time. (Plus, we're also expecting larger more powerful models in releases like GPT4).

As this happens, the "bag of tricks" prompting approaches are likely to go away.

However, the key tenant of prompt engineering will stay the same: humans learning how to communicate with AI models like GPT3, ChatGPT, DALL-E, etc.

So, should you learn prompting? It depends on your goals. The easiest approach is to wait for tools that abstract prompting away (using basically prompt templates). But if you want to be ahead or ensure you're getting the best possible results, developing your own prompts for your specific use case currently seems like the better approach.

After all, we are all basically learning how to communicate with an entirely new entity - Artificial Intelligence 🤓.

And if you're a beginner looking for ways to start learning prompt engineering, check out our free, five part "mini-course" email series, delivered daily, focused specifically on beginner prompt engineering.

We have a lot to say about prompt engineering and this is only the tip of the iceberg - so get ready for more to come. 🥶

USE CASE DEEP DIVE

Expand metadata for recommendation systems

Metadata (data that provides information about other data) can be extremely valuable. But it can be very expensive and tedious to acquire.

But one of the coolest use cases we've seen lately is using GPT to automatically generate metadata for datasets! Obviously way cheaper than sourcing it yourself.

Let's talk examples:

You have a product that recommends restaurants to consumers based on their dietary restrictions.

However, the data you have from restaurants is limited to the names of their menu items and completely lacks the ingredients used.

Enter GPT. Now you can automatically generate a first draft of the ingredients used per each restaurant menu item. From those ingredients, GPT can then guess the restaurant's dietary restriction category (gluten free, vegetarian, vegan, etc.).

With GPT, your product can help recommend restaurants to people based solely on that restaurant's menu items.

Why does this matter?

If you're trying to build any type of recommendation or filtering system, or just have a product that needs more metadata to be successful, GPT is extremely helpful in generating first drafts or estimates for your end users.

LINKS

For your reading list 📚

A look at generative AI in engineering, real estate, and education

• One of the best discussions we’ve read with multiple points of view on what could happen to software engineering roles given GPT

• In industries like real estate, the use of generative AI is becoming integral to internal workflow efficiencies

• But in education, the industry is grappling with how to respond to ChatGPT. Some teachers are adopting an open policy, while others, like New York City public schools, are completely banning the use of ChatGPT by teachers and students

And if you're really nerdy ...

• Google released its breakthrough research on its new music model: MusicLM

• Here's a paper introducing AtMan - a method to understanding transformer predictions through memory efficient attention manipulation

That's all!

We'll see you again on Thursday. Thoughts, feedback and questions are much appreciated - respond here or shoot us a note at [email protected].

Cheers,

🪄 The AI Exchange Team