We often think of privacy and discoverability as opposing forces. A conversation is either public and searchable by anyone, or hidden away in a private chat. But privacy works in more nuanced ways in our everyday lives. Think of a coffee shop - you might be deep in conversation with a friend while still overhearing snippets from the next table over. Or a conference - where people share ideas publicly, but the most real connections often happen in the sidelines between talks.
We're developing a concept called contextual privacy - or, contextual discovery. It breaks down the outdated binary of "public versus private" and instead mirrors the way privacy works in our everyday lives. How it works:
Stage 1 - Composable Privacy: With LLMs, users can now compose their context, giving them full control over what to discover. We're applying the same principle to privacy. Imagine being able to specify when you want to be discoverable.
Stage 2 - Contextual Discovery: Now imagine delegating this capability to agents. Picture every conversation as private, but with agents making it visible only to relevant people. Each message and interaction becomes a privacy setting, enabling constant negotiation and adjustment of privacy boundaries. This way, interactions are social but still contextually private, just like in real life. This dynamic approach allows for a more natural flow of information and connection online.
This breaks the binary privacy concept of the web, enabling interactions that blur the lines between public and private. Your unfinished thoughts are discoverable, and relevant strangers are joining your conversations. This new mode might initially feel strange - not knowing for whom it will be visible - but you control it. Some boundaries are more permeable, allowing more information to pass through. Similar to hotel bars: they’re public but mostly relevant people come. There’s an implicit trust mechanism at work.
By integrating these elements, we are creating a contextual environment that closely mirrors the complexity of real life social interactions. Will it become the perfect filter bubble, or a breakthrough in personalized interactions online? Either way, we’re interested in seeing what happens when privacy and discoverability coexist together.