I think it’s wise to remember generative AI as we know it today is very new. The benefit to having been an AI researcher for more than 10 years is limited.
Anyone with a good brain can join the field and make a contribution.
As for scaling and the bitter lesson, there’s no doubt NN can be trained and “learn” anything to any degree of precision. But, that’s a far cry from AGI or ASI, which is the promise that was used to raise funds.
I think 2025 is going to be challenging if the big labs don’t refocus on demonstrating what can be achieved using current models…and it’d better be profitable.
As the test-time compute silver bullet is concerned, I think it’ll stall even faster unless AI models can be put to use to achieve new breakthroughs. I say this because laying out reasoning patterns on top of deep-learned knowledge is a rail more than a launchpad.
Congrats guys :)
I think it’s wise to remember generative AI as we know it today is very new. The benefit to having been an AI researcher for more than 10 years is limited.
Anyone with a good brain can join the field and make a contribution.
As for scaling and the bitter lesson, there’s no doubt NN can be trained and “learn” anything to any degree of precision. But, that’s a far cry from AGI or ASI, which is the promise that was used to raise funds.
I think 2025 is going to be challenging if the big labs don’t refocus on demonstrating what can be achieved using current models…and it’d better be profitable.
As the test-time compute silver bullet is concerned, I think it’ll stall even faster unless AI models can be put to use to achieve new breakthroughs. I say this because laying out reasoning patterns on top of deep-learned knowledge is a rail more than a launchpad.