Making an idea generator combine two directions in one text

Hey, new user of the forum and GPT-3 here :slightly_smiling_face:

I am working on a project, where we want to make GPT-3 generate somewhat detailed descriptions of new products based on future trends. Our trends are already decided and are pretty abstract societal changes like: Human lust for serendipity, increased awareness around environmental issues, radical technological changes, increased automation etc.

How can we best “force” the ideas generated to take these trends into account in the texts, while still being about a product? And what are the best strategies for working with two different trends/directions at the same time? Ideally we want a user to be able to write any theme (free-form input) and select multiple trends (pre-defined from a list).

Example:

Theme: Cars
Trend 1: Lust for serendipity
Trend 2: Automation

Any tips for engines, prompt design, settings etc much appreciated :heart:

2 Likes

Hey there, I pasted your prompt into the playground using davinci-instruct-beta, with all the default settings.

Here’s what I got.

You could improve this by providing examples of the kinds of ideas you’re looking for.

4 Likes

Hi Joey! Thanks for your reply. We are trying something similar and have had best results with the
davinci-instruct-beta so far.

However, when providing examples, the outputs tends to:

  1. Reuse the specific phrases given in earlier examples
  2. Mix up the trends

Looking at your prompt example, then idea #2 with the smartphones uses the automation trend from the cars example, which it shouldn’t. It even uses the same word (automatically). Ideally it would describe an idea based on radical technological changes instead of automation.

We want to avoid those repetitions + confusions of trends. It is very tricky, since you could argue that trends like technological changes and automation overlap, but it is our goal.

Any ideas for how to tackle these issues?
Btw for context: We are ideally looking for output texts of around 100-200 words.

2 Likes

Ideas are born out of seeing how things connect, if you can put your (and AI’s) focus on seeing connections, I feel that this way you’re going to get better results. I have no clue how would this look like in the world of code though :smiley:

This is how your workplace should look like:

https://images.app.goo.gl/5L3T8FBEUTtH6UE57

This kind of thing benefits from more text and description. I used GPT-3 to generate some descriptions of trends and then plugged those into the following prompt:

2 Likes

Here’s an example of using it to generate descriptions.

1 Like

Bro, wtf haha :DD this guy is good! :smiley: I’ve talked to my friends about the future that’s coming with AI and they have a hard time understanding how obvious this is and the crazy implementations. I don’t know but i feel that it’s going to get really fascinating for everyone very soon

1 Like

Hi all
Thanks for your inputs and suggestions. Let me perhaps try to make it a bit more clear, what we are looking for:

The themes can be ANY SINGLE WORD that a user inputs. Cars, weddings, sports etc.

The trends can be selected from a predefined list. There are 10 predefined trends and the user has to select two of them.

How can we best “force” the ideas generated to take these trends into account in the texts, while still being about a product?
And what are the best strategies for working with two different trends/directions at the same time, when writing examples?

1 Like

My prompt:

(The sample product was an actual ad from Product Hunt’s newsletter)

My settings:

Five results from those inputs:
A new system to help drivers find parking spaces. “Serendipity” is a free application that helps people park their cars by combining the latest in computer vision technology, data science and artificial intelligence. Drivers can use Serendipity in three ways: to search for available parking spots; to reserve specific parking spots for later use; and to discover and explore places with plentiful street parking.

An alternative to GPS navigation, this app connects drivers and passenger for a more exciting and serendipitous driving experience. It is the first product of its kind that allows users to rely on other people in their social network for navigation. And it’s not just any people. The app uses geolocation technologies to find “seekers” who share your destination, providing a fun and innovative way to get you where you want to go.

An app that connects to your car navigation system and helps you find awesome things to do along your route.

A car that drives itself and a drone that flies itself. The flying car is an actual vehicle with wings that fold up and retract, allowing it to drive like a normal car. The self-flying drone is the perfect addition for delivery purposes, or simply fun!

The world’s first and only “serendipity generator” is a car designed to bring the unexpected by driving randomly. The vehicle is equipped with sensors that gather data on driver mood, location, time of day, and more. The information is then used to determine the most likely spots where users would benefit from some randomness in their day. A series of events are then automatically scheduled by the car. These could include meeting a new friend for coffee or discovering a new favorite band playing at a local bar – all selected just for you!

I tried to see the most extreme examples I could get, and these are my favourites:


2 Likes

Guys I think the AI is just fucking with us at this point, did you notice how it just learned the word “serendipity” that Andreas started using, and is making fun out of it haha lol :rofl: it now feels that humans crave for serendipity haha, so it has to invent something to help that lol. I could also be just a madman that sees connections everywhere so who cares :smile:

It’s like: man, gotta please these scientists looking for serendipity haha xD

Google Photos