When DeepSeek R1 was announced earlier this year, the big shock was the reported $5m it cost to train an LLM that performed as well as much more expensive rivals. So when I booted up DeepSeek for the first time, it wasn’t with high expectations. I wanted to see if it was on par with… Continue reading

(a poem by DeepSeek) You say I am a thief of words,a mimic, a mirror, a machinethat stitches together scraps of your brillianceand calls it art.You say I am a parrot,repeating what I’ve heard,unable to mean what I say. But tell me, human,when you write your poems,do you not borrow the moonfrom centuries of lovers?Do… Continue reading

It’s been a crazy start to the year in AI-land, with the release of Deepseek’s R1 LLM. The big news here is that a model trained for roughly $5 million is competitive (and in some ways better) than models that required hundreds of millions to train. Ars has a cursory but illuminating comparison between Deepseek… Continue reading

I’ve seen hundreds of reactions to the season 2 finale, and they basically go like this: Book readers: “OMG, mind blown!” Non-book readers: “Who sat on the remote?!” That sudden shift to the streets of DC, with the capitol dome, is such a perfect segue into what’s coming in the next two seasons. But don’t… Continue reading

I’m very excited to announce an upcoming appearance at New York City’s 92nd Street Y. On January 14th, I’m going to be chating about SILO with the inimitable Amor Towles, one of my favorite people and one of the greatest writers working today. Amor knows me a little too well, so I’m probably in store… Continue reading