One of Rusty's all time best, well worth your time: https://www.todayintabs.com/p/a-i-isn-t-people
-
One of Rusty's all time best, well worth your time: https://www.todayintabs.com/p/a-i-isn-t-people
-
One of Rusty's all time best, well worth your time: https://www.todayintabs.com/p/a-i-isn-t-people
@codinghorror @temptoetiam
“Saying that any of this is teaching a chatbot philosophy is like calling yourself a psychiatrist because you thought about Oedipus while jacking off. “🧑
-
One of Rusty's all time best, well worth your time: https://www.todayintabs.com/p/a-i-isn-t-people
@codinghorror “Imagine you have two machines. One you can open up and examine all of its workings, and if you give it every picture of a cat on the whole internet, it can reliably distinguish cats from non-cats. The other is a black box and it can also reliably distinguish cats from non-cats if you give it half a dozen pictures of cats, some apple sauce, and a hug. These machines sort of do the same thing, but even without knowing how the second one works I am extremely confident in saying it doesn’t work the same way as the first one.”
-
R relay@relay.an.exchange shared this topic
-
One of Rusty's all time best, well worth your time: https://www.todayintabs.com/p/a-i-isn-t-people
Thank you - still reading. But, this:
" Andrej Karpathy recently implemented a fully functional large language model in 200 lines of Python code. He also wrote a detailed explanation of exactly what it does, how it works, and the ways it differs from big company frontier models, which are all just differences of scale and efficiency."
-
One of Rusty's all time best, well worth your time: https://www.todayintabs.com/p/a-i-isn-t-people
@codinghorror If it was just cats it would be one thing, but having non-deterministic systems identify tumors or drones is terrifying
-
One of Rusty's all time best, well worth your time: https://www.todayintabs.com/p/a-i-isn-t-people
@codinghorror I think there are a bunch of terrible arguments in that essay. Focusing on the 200 lines of Python is like saying “meat that thinks?!” Everyone thinks human intelligence is so special and I think we’re going to come out of this in 5 or 10 or 20 years and conclude that it isn’t as special as we thought, just like Life wasn’t due to a special life force and souls don’t exist. 1/
-
@codinghorror I think there are a bunch of terrible arguments in that essay. Focusing on the 200 lines of Python is like saying “meat that thinks?!” Everyone thinks human intelligence is so special and I think we’re going to come out of this in 5 or 10 or 20 years and conclude that it isn’t as special as we thought, just like Life wasn’t due to a special life force and souls don’t exist. 1/
@codinghorror The question is not what the 200 lines of code do, they are merely like the handful of neurotransmitters in out biology. The question is what’s going on in the billions of model parameters that seem to encode not only word patterns but also some algorithms for behavior.
The issue of needing to see more training examples than humans need has been addressed by pointing to the work evolution has done in shaping humans that we need to do brute force for LLMs. 2/
-
@codinghorror The question is not what the 200 lines of code do, they are merely like the handful of neurotransmitters in out biology. The question is what’s going on in the billions of model parameters that seem to encode not only word patterns but also some algorithms for behavior.
The issue of needing to see more training examples than humans need has been addressed by pointing to the work evolution has done in shaping humans that we need to do brute force for LLMs. 2/
@codinghorror And yes, there *are* issues with lack of grounding in the physical world.
I don’t think LLMs are synthetic humans or have emotions, but once we run these things with persistent state and long continuous loops, I expect the results to start to resemble humans a lot more.
-
One of Rusty's all time best, well worth your time: https://www.todayintabs.com/p/a-i-isn-t-people
@codinghorror best point is right away at the start. Intelligence does not need "all the data" to learn.
-
One of Rusty's all time best, well worth your time: https://www.todayintabs.com/p/a-i-isn-t-people
@codinghorror ai is people the question is what are they going to do?
#techcrunch.com
-
R relay@relay.infosec.exchange shared this topic
