The New 'Transparent' Twitter Isn't Very Transparent
twitteralgorithmic-transparencyelon-muskrecommendation-systemsopen-source
Abstraction: Critique of Twitter's partial algorithm open-sourcing as superficial transparency theater
Key points:
- Twitter released source code for its "For You" recommendation algorithm on GitHub in April 2023 under Elon Musk's "new era of transparency" branding
- Researchers call the code "pretty standard" (Zachary Steinert-Threlkeld, UCLA); it omits the underlying AI models and their training data
- System pulls 1,500 tweets from hundreds of millions for a user's feed; photos/videos get a ranking boost; likes may boost visibility more than replies
- Musk's account has a special hard-coded boost in the algorithm, seemingly confirming earlier reports of engineer pressure
- Twitter simultaneously raised researcher API access fees to $42,000–$210,000/month, undermining independent study
- Critics (Center for Countering Digital Hate, Algorithmic Transparency Institute) argue Twitter became less transparent overall under Musk despite the gesture
Connections: Twitter · Elon Musk · Algorithmic Transparency · Recommendation Systems · Content Moderation