Poetry in SF: Summer 2026
This summer, I co-hosted a few poetry gatherings in San Francisco, bringing together friends from Trajectory, Physical Intelligence, and Inception Labs.
In these accelerationist times, poetry is a uniquely interesting medium to study with a techie/fuzzy lens. Consider two translations of 2011 Nobel laureate Tomas Tranströmer:
During our guided discussion, we pored over what endows each version with poetic (vs. prose) qualities—conjunction omission, verb choice, etc. Yet the "so what" of all these devices converges with W.S. Merwin's description that "[a] poem begins to be a poem when a sequence of words starts giving off what you might describe as a kind of electric charge, when it begins to have a life of its own."
With much focus today on optimizing the semantic ability of LLMs (whether tokenmaxxing or valuemaxxing), it is thought-provoking to engage with poetry, where the next "token" is chosen for what it does rather than what it means. Every candidate is defensible on meaning; what separates them is qualia, the jolt of "deleted" after the bus has been transformed from a vehicle to an "it." 1,2
It is my hope that interest in poetry will increase as we continue to make progress in AI. If you are interested in future poetry events (SF and NYC), please reach out!
I suspect this is also why AI music won't land the same as a song by an artist whose lore one has been tracking for a decade (IYKYK), even though AI can replicate the ice cream progression. Case in point ↩
This is why "skill issue, turn up the temperature" doesn't solve, or at the very least misses the point. Temperature is rank-preserving, changing how boldly a model samples, never what it prefers or feels. ↩