Show HN: GPT-3 generated articles about AI

Here's a fun little project I did today on a whim. I happen to have access to the OpenAI API, so I used their playground feature to generate AI headlines with their taglines. I fed it this prompt:

'Last week's top AI news: * Caltech unit creates AI helping drones to withstand violent winds - "Caltech researchers are developing a drone with rapidly reacting artificial intelligence (AI) capacities that allow it to adapt in flight to extreme wind similar to tornado or hurricane conditions."

* How Deep Squeak, an AI program with a weird name, is detecting whales - "Artificial Intelligence is booming. And now an AI program is being used to search for whales." "

Plug alert: this text sourced from the latest edition of the Last Week in AI newsletter (https://lastweekin.ai/) which I co-run. So here's what it got just generating text given that prompt:

* Google’s DeepMind creates AI that can beat humans at hide-and-seek - "A new artificial intelligence program developed by Google’s DeepMind can beat humans at hide-and-seek, according to a new paper."

* Carnegie Mellon Researchers Develop AI System That Can Help You Write Better Emails - "Researchers at Carnegie Mellon University have developed an artificial intelligence system that can help you write better emails."

* New AI System Can Help You Find a Job You’ll Love - "A new artificial intelligence system developed by researchers at the University of Toronto can help you find a job you’ll love, according to a new study."

Perhaps unsurprisingly, quite good. Though, I was surprised to find it repeats itself super quickly. Would have thought GPT-3 had solved this very basic language model problem... It also seems to repeat actual news stories from the past, such as:

* Google DeepMind's AlphaGo Zero AI can teach itself Go and other games in hours - "DeepMind has created an artificial intelligence system that can not only defeat humans at the game of Go, but also teach itself to play from scratch within hours."

I also tried prompting it with just "Last week's top AI news" and got the following:

1. Google's DeepMind has created a new algorithm that can predict how proteins will fold.

2. Facebook's artificial intelligence research lab has developed a new system that can automatically generate 3D models of objects from 2D images.

3. IBM has announced that its Watson AI platform will now be available to developers on the IBM Cloud.

Again, repeating facts from the past. Lastly, how about "This week's trending AI papers" as a prompt. Also not bad:

1. "Deep Learning for Recommender Systems"

2. "A Neural Network Approach to Context-Aware Query Suggestion"

3. "Learning Deep Representations of Fine-Grained Visual Concepts"

A while back I fine-tuned GPT-2 on 100 weeks' worth of AI news (and released the dataset+colab notebook, see https://www.skynettoday.com/digests/ai-news-analysis), and got some comparatively more amusing stuff:

"Researchers develop a traffic light for self-driving cars - A pair of new papers from University of Tokyo Institute of Technology researchers and the World Health Organization claim to have addressed one of the great challenges of artificial intelligence: traffic lights.

How I used NLP's GPT-3 to write the AI I Created - When it comes to creating AI, there are 2 main choices I have:- write a clean code or- just use the output from the original app to clean code.

Meet Microsoft's first AI-powered coffee machine - Jeff Dean, chair of Microsoft’s artificial intelligence (AI) division, recently sat down with WIRED senior writer Will Knight to discuss the value of building products that do a lot of good, but don’t always do much of anything else."

That's about it, nothing too fancy but kind of fun. Feel free to suggest other little experiments to try!