Tech Brew Ride Home Mon. 08/05 – OpenAI Can Detect AI Text But Isn’t Sure It Wants To
Aug 5, 2024
OpenAI is grappling with the launch of a promising tool to detect AI-generated text but is hesitant due to potential backlash. The correlation between Bitcoin and traditional markets raises eyebrows as it dips below $50,000. Elon Musk is reigniting his lawsuit against OpenAI, adding another twist to the tale. Meanwhile, China is gearing up to challenge Starlink with its own satellite constellation. Lastly, the music industry faces uncertainty as major players struggle with slowing streaming growth.
AI Snips
Chapters
Transcript
Episode notes
OpenAI's AI Detection Dilemma
- OpenAI has a 99.9% reliable AI text detector but hasn't released it.
- They are balancing transparency with user retention, as some users would be turned off by it.
How OpenAI's AI Detection Works
- OpenAI's AI detector uses watermarking, subtly changing word selection to leave a detectable pattern.
- This method is highly effective, but some worry it can be circumvented by simple techniques like translation or adding emojis.
Access and Priorities for AI Detection
- OpenAI grapples with who should access the detector: too few limits usefulness, too many risk reverse-engineering.
- They've prioritized image and audio watermarking over text due to greater potential harms, especially during election years.