New side hustle alert: teaching ChatGPT how to be a better writer.
Driving the news: Companies are paying people to make content solely to train AI. That includes hiring people with writing experience and graduate degrees for part-time essay-writing gigs, or Adobe offering about $3 per minute for videos of people doing everyday tasks.
Catch-up: AI companies are willing to pay up for training data that will make their models smarter than the competition. Major players like OpenAI, Google, and Apple have pursued deals to license content from the likes of Reddit, Tumblr, Shutterstock, and news outlets.
- Licensing data from platforms is legal, but it doesn’t mean users will be happy about it — did you know that Photobucket is shopping around your 15-year-old uploads?
Why it matters: Using content created from scratch is a reversal of how most AI models have been trained — automatically scraping data from across the internet. Paying people to make original content not only gives a platform data others don’t have, but avoids copyright issues AI has been criticized for.
- This could also be a form of quality control, since AI models don’t always scrape data from the best sources, and often just copies content made by a different AI.
Zoom out: Being copyright friendly is a chance to stand out in the eyes of AI’s most sought-after paying customers. Creatives are obvious users for video and photo tools, but are resistant to AI that was trained without permission on their work. Enterprises might also slow down adoption due to copyright issues.