Hi HN, I'm the author. I recently finished a draft of a book on the political economy of AI, arguing that our main economic bottleneck is shifting from generating information to verifying it.
This essay is a spin-off from that research. It’s about 'stranded knowledge' — research that is mathematically or scientifically sound but gets buried because it lacks institutional legitimacy at the time (like Hochreiter’s 1991 thesis before LSTM took off).
My argument is that as AI drives the cost of generating new papers to near zero, the archive decays 'attentionally'. I think the highest-return use of AI right now isn't writing more papers, but building a triage system to recover the neglected ones.
I’d love to hear your thoughts, especially from anyone working on metascience or AI discovery tools.
Hi HN, I'm the author. I recently finished a draft of a book on the political economy of AI, arguing that our main economic bottleneck is shifting from generating information to verifying it.
This essay is a spin-off from that research. It’s about 'stranded knowledge' — research that is mathematically or scientifically sound but gets buried because it lacks institutional legitimacy at the time (like Hochreiter’s 1991 thesis before LSTM took off).
My argument is that as AI drives the cost of generating new papers to near zero, the archive decays 'attentionally'. I think the highest-return use of AI right now isn't writing more papers, but building a triage system to recover the neglected ones.
I’d love to hear your thoughts, especially from anyone working on metascience or AI discovery tools.