Great read. Even in STEM research as a grad student I’m very tired of every saying “let’s try machine learning on this problem” to get something that works marginally better than some conventional models but requiring huge amounts of computation and data.
I work professionally with actually useful ML stuff (we parse a lot of weird ass files and it’s extremely powerful in that context) - we’ve looked at integrating gpt3 and it scored much worse on accuracy than the model we trained in-house. We’re also investigating adding front-end AI bullshit to placate the CEO. Even at the good shops, you’ll probably get buried in this bullshit - but there are good opportunities out there!
Great read. Even in STEM research as a grad student I’m very tired of every saying “let’s try machine learning on this problem” to get something that works marginally better than some conventional models but requiring huge amounts of computation and data.
I work professionally with actually useful ML stuff (we parse a lot of weird ass files and it’s extremely powerful in that context) - we’ve looked at integrating gpt3 and it scored much worse on accuracy than the model we trained in-house. We’re also investigating adding front-end AI bullshit to placate the CEO. Even at the good shops, you’ll probably get buried in this bullshit - but there are good opportunities out there!