Reddit initially tested words like "interesting/uninteresting" for its voting system but found they imposed too much specific meaning. For example, a user might not want to label a story about a natural disaster as "cool." Ambiguous up/down arrows allow for intuitive, context-dependent meaning, a key lesson in universal product design.

Reddit does the opposite of every other social platform (e.g., anonymity, text-first). This wasn't initially a conscious business strategy but a result of being stubborn about their core values and what felt right for their product. This "stubbornness" organically created a powerful, differentiated position in the market.

When a norm is violated publicly, it threatens the common knowledge that the norm exists and is enforced. The resulting public punishment, like a digital-age pillory, isn't just about the transgressor; it's a signal to the entire community that the norm is still in effect, thereby restoring common knowledge.

Reddit consciously avoided growth tactics like promoting "enraging" content that made competitors explode. This was a values-driven business decision that meant slower growth but preserved the platform's core authenticity, which has now become its key differentiator in the social media landscape.

Social media content that "dunks on" an opposing group is 67% more likely to be shared. This virality is driven by in-group reinforcement, not by persuading outsiders. The platform's algorithm rewards and encourages this divisive behavior.

According to CEO Steve Huffman, almost every policy change at Reddit begins with a gut feeling that "something feels off" or "morally not right." The team's primary job is then to work backward from that instinct and articulate it into a coherent, enforceable policy. The intuition comes long before the formal rule.

Treat your online community like a dinner party at your home. To prevent toxicity, immediately remove members who commit even minor rule violations. This 'broken windows' approach stops the standard of acceptable behavior from degrading over time.

Steve Huffman previously thought AI could never work for moderation because it lacked context. Now, he believes large language models (LLMs) are powerful enough to understand nuance like humor and sarcasm. The future of Reddit moderation is human moderators writing the rules, but AI enforcing them at scale, a dramatic shift in thinking.

The metric for a successful community has shifted from high activity ("noise") to high trust. Members no longer want to sift through hundreds of discussions. They want a smaller, curated space where they can trust the expertise and intentions of the other people in the room.