I’ve been thinking lately about using AI to aggregate and respond to current/contextual information as it seems to me that’s a better use of it than something as general as Web search. With Web search the AI requires a lot of contextual understanding and common sense which it doesn’t have, so you end up with search results about glue on pizza and so forth. But if you give an AI a brief, well-defined prompt with lots of supporting content, it can work with that in interesting ways.

To explore the idea a bit I made a program called Collab-O-Matic, which uses AI to generate YouTuber collaboration ideas.

The program starts by grabbing two random YouTubers from the Wikipedia list and getting their respective articles. To add more current data, Collab-O-Matic also pulls the YouTube RSS feeds of each YouTuber (if they have Wikidata property P2397 available.) An Anthropic API call then digests all that information and uses it to suggest a collaboration between the two YouTubers. Sometimes, it refuses based on the content of one of the selections or because the two YouTubers are too different (a toy reviewing channel paired with a firearms channel, for example.)

I’m including the suggested collaboration between Tim Chantarangsu and Casey Neistat below. It’s not as compact as just getting it in a screenshot, but it seems to me a bit more accessibility-friendly (if I am incorrect I am open to be educated.)

Your Next Viral Collab!

Here’s a creative idea for a hypothetical collaboration video between Tim Chantarangsu and Casey Neistat:

Title: “NYC’s Wildest Food Challenges ft. Casey Neistat – Can We Conquer Them All in 24 Hours?”

Video Format and Structure:
Tim and Casey team up to take on some of the most extreme, unique and delicious food challenges across New York City, all within a 24 hour period. They’ll visit a variety of establishments, from hole-in-the-wall joints to upscale eateries, attempting challenges like super spicy wings, gargantuan pizza slices, towering ice cream sundaes, and more. The video will be shot in Casey’s signature vlog style, with Tim adding his comedic commentary and food expertise throughout.

The 24 hour time limit adds an element of suspense and urgency. Between challenges, Casey and Tim will explore the city, sharing stories and reflecting on the food adventure. The video will culminate with a final challenge at a famous NYC restaurant, where they must consume a 5 course tasting menu in record time.

Unique Contributions:
Casey’s cinematography skills and ability to make everyday experiences feel epic will elevate the video’s production value. His knowledge of New York will also help uncover some hidden gem eateries.

Tim’s background as a food personality and comedian will keep the video entertaining, insightful and light-hearted. He can provide fun facts about the foods they’re eating and humorous reactions to the intense challenges.

Audience Engagement:
Combined, Tim and Casey’s audiences will enjoy seeing these two dynamic personalities explore NYC’s food scene together. The video’s competitive aspect, unique foods, and ticking clock will keep viewers engaged, rooting for the duo to successfully complete all the challenges.

Holiday Tie-In:
With the International Day of Persons with Disabilities on December 3rd, Tim and Casey could donate a portion of the video’s proceeds to a charity supporting this cause. They could also give a shoutout to disability organization and share how food can bring people together.

This collaboration would blend food, comedy, adventure and heart – signature elements both fanbases know and love. Seeing Tim and Casey learn from each other and forge a friendship over wild food challenges is sure to satisfy viewers.

That worked pretty well, but sometimes the suggestions got a bit generic; a tech channel and a music channel reviewed a lot of music tech. So to give it another dimension I added in the API Ninjas’ holidays API and tossed into the Anthropic prompt any holidays coming up in the next 30 days. As you can see it actually picked up on the International Day of Persons with Disabilities on December 3 in the example above, though a lot of the time the suggested tie-ins were more around major winter holidays. I expected that to be somewhat limited in what it would add to the collaboration suggestions, but I was surprised at how much of a dimension it added — holiday songs, gift exchanges, and even things like sampling weird eggnog flavors and joke gifts have come up in the AI-generated scenarios. 

This was loads of stupid fun but I was still missing one of the most essential aspects of YouTube: the over the top thumbnail. So I added in a DALL-E call to generate a thumbnail for each suggested scenario.
In the case of the Tim/Casey suggestion it generated… this. I don’t know much about those YouTubers, but I would guess this isn’t them in the picture.

The thumbnails have varied a lot in their quality and their relevance and most of all their spelling, and I can’t play with this too program much because APIs are expensive, but I have laughed a lot today at some of these creations. 🤣