I just took the time to read most his 2022 post and it didnt hold up very well. He illustrates why gpt 3 is a dead end but I just tried the exact example on gpt 4o and it passed with flying colors.
The premise is that scaling wonâ(TM)t work and that the world is overly focused on the LLM approach.
He claims tagging images wont work well enough for radiology. Fine but companies like Surona medical seem to be still around years later.
He claims that image tagging wont work but an iPhone does an amazing job of identifying species of things in nature⦠absolutely amazing. And you can now generate captions for image locally on an old MacBook using hugging face models. Itâ(TM)s not perfect but I can be sure that itâ(TM)s made a ton of progress since 2022. So the world isnt just focused on llms these are computer vision models.
Suno is generating mediocre music, far better than it did last year when it was abysmal and some music AI didnt even have language two years ago. Now you can generate country music about retiring your toothbrush that sounds legit.
Finally computer coding is obviously working quite well.
He makes the old tripe of the probabilistic stochastic parrot not having comprehension. Sorry but I believe thats how the human brain works as well.