I share your frustration at those who confidently & prematurely write-off rapidly-changing AI tech based on dated examples, cherry-picked anecdotes from the unskilled, & zero extrapolation based on momentum. They do a a double-disservice to those who trust them: 1st, by discouraging beneficial work on ripe, solvable challenges, and 2nd, encouraging a complacency about rapid new capabilities that may leave vulnerable people at the mercy of others who were better prepared.
But, not being familiar with the account in question, I don't see those attitudes in that tweet. It seems more an assessment "no one has quite nailed this yet" than defeatism over whether it's possible.
> i have yet to see any ai system properly implement individual level-adjusting tutoring. i suspect because the LLM needs a proper theory of mind (https://twitter.com/swyx/status/1697121327143150004) before you can put this to practice.
But to be perfectly transparent, I'd never respond so harshly to someone for just that tweet, or even that comment.
Instead it's the fact they're currently a synecdoche for the crypto-ization of AI. This person doesn't usually dismiss AI, instead they heavily amplify the least helpful interpretations of it.
_
This is one of the largest voices behind the new "the rise of the AI engineer" movement in which this author specifically claimed researchers were now obsolete to AI due to the tooling they built: https://news.ycombinator.com/item?id=36538423
Like, I get wanting to make money by capturing value as much as the next person... but basing an entire brand on declaring that the people who are enabling your value proposition are irrelevant just to create a name for yourself is pointlessly distasteful.
The only thing he gained by saying researchers don't matter and understanding Attention doesn't matter is exactly I described above: a wild opinion that attracted the unsure, pissed off the knowledgeable, and served as a wedge that he could then carve out increasingly large slices of the pie for himself with.
Fast forward 2 months and now the process has done its thing, the "AI engineer" conference is being sponsored by the research driven orgs because they don't want to be on the wrong side of the steamroller.
I share your frustration at those who confidently & prematurely write-off rapidly-changing AI tech based on dated examples, cherry-picked anecdotes from the unskilled, & zero extrapolation based on momentum. They do a a double-disservice to those who trust them: 1st, by discouraging beneficial work on ripe, solvable challenges, and 2nd, encouraging a complacency about rapid new capabilities that may leave vulnerable people at the mercy of others who were better prepared.
But, not being familiar with the account in question, I don't see those attitudes in that tweet. It seems more an assessment "no one has quite nailed this yet" than defeatism over whether it's possible.