The fascinating thing here that I think caught many of us by surprise— and certainly myself — is realizing that building capable LLMs so far is essentially elas

The fascinating thing here that I think caught many of us by surprise— and certainly myself — is realizing that building capable LLMs so far is essentially elastic with respect to capital. I don’t think this was immediately obvious when OpenAI first launched ChatGPT. I also don’t think this is true in general for software (though there are other cases).