4 ms·

John Carmack's Approach to AGI

45 points by carabiner 4y ago

zora_goron 4y ago
HN discussion on his full podcast with Lex Fridman - https://news.ycombinator.com/item?id=32347172 https://news.ycombinator.com/item?id=32347172 (which is where these comments are from)
MikeDelta 4y ago
AGI is Artificial General Intelligence, for those who were wondering.
cheeselip420 4y ago
Carmack is a genius, but I somehow doubt that is going to be the one to crack AGI... I can't wait to see what we learn from him though.
- JonChesterfield 4y ago
  I'm not going to bet against him on this one.
mountainriver 4y ago
Cool he’s giving it a spin but AGI is its own beast and as a novice on the subject his predictions are probably not well founded
- djmips 4y ago
  Agreed, look at how optimistic he was at the time of his rocket science endeavors.
treebeard901 4y ago
His role is likely more focused on optimizing the GPU usage, which for AGI will be the largest upfront cost. His work optimizing Oculus and ID tech engines all those years will be very valuable. He is from the old school programmer philosophy that required code to be as efficient and creative as possible to even run on the hardware of the day. There are fewer and fewer opportunties to have rewarding work like that since more compute can just be added instead of spending the hours refactoring slow code. Ah, the good ole days when the relationship between hardware and software wasnt given an easy mode for developing new software at the expense of performance. Kids these days dont know how good they have it. Back in my day, every clock counted! End rant
- djmips 4y ago
  Don't worry those days are still here. Video game coders have to contend with 8 year old console hardware. And I predict this will seep into all aspects of programming as Moores law continues to arrest.
- cheeselip420 4y ago
  he's not claiming that he just needs to help it scale up. He's claiming that there are a few key insights missing between today's state of the art and AGI. I mean, he may be correct about there only being a few more steps required, but I still suspect he's overestimating his own capabilities while underestimating the scope of the problem.

45 points by carabiner 4y ago

HN discussion on his full podcast with Lex Fridman - https://news.ycombinator.com/item?id=32347172 https://news.ycombinator.com/item?id=32347172 (which is where these comments are from)

MikeDelta 4y ago

AGI is Artificial General Intelligence, for those who were wondering.

cheeselip420 4y ago

Carmack is a genius, but I somehow doubt that is going to be the one to crack AGI... I can't wait to see what we learn from him though.

JonChesterfield 4y ago
I'm not going to bet against him on this one.

mountainriver 4y ago

Cool he’s giving it a spin but AGI is its own beast and as a novice on the subject his predictions are probably not well founded

djmips 4y ago
Agreed, look at how optimistic he was at the time of his rocket science endeavors.

treebeard901 4y ago

His role is likely more focused on optimizing the GPU usage, which for AGI will be the largest upfront cost. His work optimizing Oculus and ID tech engines all those years will be very valuable. He is from the old school programmer philosophy that required code to be as efficient and creative as possible to even run on the hardware of the day. There are fewer and fewer opportunties to have rewarding work like that since more compute can just be added instead of spending the hours refactoring slow code. Ah, the good ole days when the relationship between hardware and software wasnt given an easy mode for developing new software at the expense of performance. Kids these days dont know how good they have it. Back in my day, every clock counted! End rant

djmips 4y ago
Don't worry those days are still here. Video game coders have to contend with 8 year old console hardware. And I predict this will seep into all aspects of programming as Moores law continues to arrest.
cheeselip420 4y ago
he's not claiming that he just needs to help it scale up. He's claiming that there are a few key insights missing between today's state of the art and AGI. I mean, he may be correct about there only being a few more steps required, but I still suspect he's overestimating his own capabilities while underestimating the scope of the problem.