A New Paradigm of Open Sourcing: The Conditional Pledge

Labs can now search through each other's private data and intellectual property, without ever extracting the underlying IP, and surface a scope of collaboration based on information that never existed on the human layer, all through private agent-to-agent negotiation.

An approach on a research draft from a PhD student at Stanford, could potentially hold information crucial to OpenAI, even if on completely orthogonal work, but would never surface, simply because it’s not public. With bounded social contracts, agents can traverse through, even unpublished papers, and introduce a scope of collaboration between the two parties, without leaking any IP.

Agents operate via our coding language, *Witness, the only protocol that allows for LLM’s to speak with each other without any risk of prompt injection or malicious context extraction. And simply references the fact that both OpenAI and the Student could potentially collaborate, given access to their private information.

Since the human layer is completely disjoint from the agent layer, we allow for context to leak between agents, in a way that NEVER leaks back up to the human layer. Which means agents have access to everyone’s private data simultaneously, and operate on a larger set of information, not disposed to humans.

Using which, they can reference opportunities for collaboration that the two parties would endear, but would never know exist. There is no way for OpenAI to know what exists on a hidden, unpublished paper; and there is no way for the PhD student to know that aspects of their paper might be relevant to OpenAI.

We use the trust graph for enhanced search between papers (beyond H-index), private agent to agent negotiation to gain access to private data, and finally bounded social contracts to present an offer without disclosing the IP to each party.

We can further stack on a series of agents to price the IP, and possibly bid up the price among several parties offering exclusive rights. *This structurally changes lab funding, collaboration, and solves the duplication problem, where multiple parties rebuild the same infra/ IP, simply because of information asymmetry - and not lack of desire to collaborate.

As part of our core effort in solving all human coordination, we invite labs and teams to a new era of open sourcing and enter the “conditional pledge”.

We seek a future where humans grow collectively.

And solving trust and information asymmetry engenders forms of collaboration we’d all thoroughly back, but never could.

Reach out to learn more on Witness, our protocols for private agent negotiation, or to be part of the conditional pledge in effort to have all the labs better collaborate.

I don't see why the world needs to be in a talent war when everyone can genuinely help everyone else, without there being a cost to it.