Great read. Particularly love the point about the ‘short dumb money’ trade being to invest earlier and take advantage of the distorted cost of capital that comes in at later stages. Very cool
Fantastic analysis as always. What would the VC market charts look like without AI deals? Now is a great time to find hidden gems in less buzzy sectors that aren't receiving as much attention from dumb money.
The need to discretely retrain the models periodically seems like something that could be better handled by making smaller models that are more specialized (e.g. a Medical LLM rather than ChatGPT 4.0) where you can refresh for newer information with less cost and fewer risks of Catastrophic interference. Am I off base here?
*Maybe* the world looks like small models which are up to date and big frontier models that are slightly more stale.
But it could be that even if inference becomes more diversified (i.e. split amongst many different models of different size), all of the models need to be semi-up-to-date which would still necessitate re-training your large frontier model every so often.
Great read. Particularly love the point about the ‘short dumb money’ trade being to invest earlier and take advantage of the distorted cost of capital that comes in at later stages. Very cool
Fantastic analysis as always. What would the VC market charts look like without AI deals? Now is a great time to find hidden gems in less buzzy sectors that aren't receiving as much attention from dumb money.
Ya. Am particularly interested in bio and healthcare services which are unloved
*healthcare services*
The need to discretely retrain the models periodically seems like something that could be better handled by making smaller models that are more specialized (e.g. a Medical LLM rather than ChatGPT 4.0) where you can refresh for newer information with less cost and fewer risks of Catastrophic interference. Am I off base here?
*Maybe* the world looks like small models which are up to date and big frontier models that are slightly more stale.
But it could be that even if inference becomes more diversified (i.e. split amongst many different models of different size), all of the models need to be semi-up-to-date which would still necessitate re-training your large frontier model every so often.