An Age of Secrets
The current internet is characterized in part by a culture of substantial openness. Information wants to be free. Ideas are a dime a dozen, execution is what matters. Sunshine is the best disinfectant. Most concretely: virtually every experience I might want to know about (ex: what’s the second trapeze class like?) has been explored on reddit. Arxiv seems to be the dominant mode of sharing academic work (to my non-academic eye).
There’s obviously substantial value to sharing information! At least part of that value seems to come from the benefit of having more people work on a problem. In my own current day to day work on verifiable inference for LLMs, my instincts are less about “how do we build this ourselves” and more “how do we make it more likely that the wider world can achieve this goal?”
On the other hand, the world we are heading toward seems to be one where intelligence is free and unlimited. One complement to intelligence is information. It’s not the only one, and it might not even be the most important one, but the implication of this is that information that cannot be derived from some public source ought to go up in value.
This is even more true in a world of agentic intelligence, where the ability to execute is also far higher. Suddenly, secret information about a particular industry is a primary differentiator–not the ability to execute. (Caveat: connections and relationships are also relevant.) This suggests that domain knowledge and contingent specifics of a niche are even more valuable–both to discover and to defend.
One of my cousins is a podiatrist–he’s long had a dream of building an app for podiatrists, but of course he’s not going to do that! It requires engineering expertise. But now that LLMs can go and do that, it’s plausible that he could go and create this app–and since not that many people actually know the arcana of foot medicine (and what podiatrists need that they don’t currently get), that’s now a thing that could happen. Previously, his move would need to be: get someone excited enough to work on the idea. Share it! They weren’t going to steal it. Now, his move could be: do it with Claude. The value of sharing information is … different.
I don’t think it’s unilaterally true that information won’t be shared in the future–there are many places where sharing information has positive sum benefits, particularly when the value of new insights or breakthroughs would be widely distributed and it’s difficult to imagine any one person being capable alone (even with LLM support) of finding the insight.
If your odds of finding the key patentable idea in a domain are 1 in a million, and the benefits you can capture are X (factoring in costs), then you only need the idea to provide you X/1 million in value to prefer making it easier for the world to find the idea than to capture it yourself.
But in a lot of cases, the odds of finding that idea are much higher, and with agentic intelligence it becomes much easier to exploit that idea in the first place (i.e X goes up because cost goes down).
In this regard, it seems to me that ironically, in an era where intelligence becomes too cheap to meter, information may become too valuable to share. Perhaps we will enter an age of secrets.