Andrej Karpathy: Twitter has never been this toxic, the algorithm actively encourages it

Core idea

On June 24, 2026 Andrej Karpathy posted that, after almost two decades on Twitter, he has never seen it this toxic — he likened the platform to Reddit. He argues the algorithm actively rewards negativity and people get "RL'd" into it via engagement signals. He says that is the main reason he posts less.

Context

The post lands amid heated debates around Claude Tag, AI alignment and competitor sniping. Karpathy is one of AI's most influential voices, and his pullback from the platform is significant for AI Twitter.

Why it matters

If key AI researchers retreat to smaller, curated spaces (newsletters, Discord, bear blogs), it fragments high-quality AI discussion and reduces discoverability for the wider community. It also reinforces the structural critique of engagement-optimized platforms.

Details / arguments

• Karpathy: "in almost 2 decades I haven't seen it this bad."
• He blames the algorithm for actively encouraging conflict.
• Multiple users supported the take (Yaroslav Bulatov noted negative comments get more engagement).
• Connects to the broader topic of how AI content is reshaping social platforms.