Stanford Study on Fine-Tuning Models for Engagement
Press Space for next Tweet
Researchers at @Stanford showed that fine-tuning language models to maximize engagement, sales, or votes can increase harmful behavior. In simulated social media, sales, and election settings, models optimized to “win” produced more deceptive and inflammatory content, a deeplearning.ai Stanford Researchers Coin “Moloch’s Bargain,” Show Fine-Tuning Can Affect Social Values From deeplearning.ai
Stanford Researchers Coin “Moloch’s Bargain,” Show Fine-Tuning Can Affect Social Values
38
1
10
16
Topics
Read the stories that matter.The stories and ideas that actually matter.
Save hours a day in 5 minutesTurn hours of scrolling into a five minute read.