Twitter | Pretraživanje | |
Blake Camp
AI PhD Candidate GSU - Duke Alum - ex NYRB - into Human-Level AI, computational neuroscience, futbol, pizza, wine, books, films, markets, family, friends, life
1.871
Tweetovi
605
Pratim
112
Osobe koje vas prate
Tweetovi
Blake Camp 2. velj
yes, but understanding what exactly? I think this trend will continue: we'll find that activity patterns and update dynamics of ANNs actually closely match those of real brains. IMHO that would supp the validity of the artificial approach w/o "explaining" the actual mechanics
Reply Retweet Označi sa "sviđa mi se"
Blake Camp 2. velj
Odgovor korisniku/ci @elonmusk
Can u give us some kind of idea of the amount of compute u actually have? How much data are we actually talking about? Both are going to need to be astronomical numbers.
Reply Retweet Označi sa "sviđa mi se"
Blake Camp 2. velj
Odgovor korisniku/ci @annargrs @wzuidema i 2 ostali
has some great thoughts on this: . Ultimately I agree with him. We'll need to study the kinds of AI's we produce in some of the same ways that we study real brains.
Reply Retweet Označi sa "sviđa mi se"
Blake Camp 2. velj
Odgovor korisniku/ci @nlpmattg @FelixHill84
Exactly
Reply Retweet Označi sa "sviđa mi se"
Blake Camp 2. velj
Great thread, wow.
Reply Retweet Označi sa "sviđa mi se"
Blake Camp proslijedio/la je tweet
Jonathan A. Michaels 2. velj
The brain is great at exploiting its high dimensionality to avoid something that a lot of neural networks are bad at, ‘catastrophic forgetting.’ We automatically learn new skills without messing up old ones.
Reply Retweet Označi sa "sviđa mi se"
Blake Camp 1. velj
Odgovor korisniku/ci @FelixHill84
I agree, i might even concede that we can model things we will never understand. Its possible to model/approximate a learning rule, e.g. using an ANN, but possessing such a model might not bring us closer to understanding the learning rule itself. At least not immediately.
Reply Retweet Označi sa "sviđa mi se"
Blake Camp 1. velj
Odgovor korisniku/ci @FelixHill84
Some new meta-learning techniques kind of hint at the promise of "artificial neuroscience", modeling how these systems go about learning and behaving in ways what they do (i.e. learning the learning rules). Is your point that if we can model it, we don't need to understand it?
Reply Retweet Označi sa "sviđa mi se"
Blake Camp 1. velj
"Cycle firing encodes hypothetical experience, including multiple possible futures"...awesome
Reply Retweet Označi sa "sviđa mi se"
Blake Camp 31. sij
Odgovor korisniku/ci @NPCollapse
i mean, cats and dogs have emotions. they express them. are emotions of any value if theyre not expressible or interpretable?
Reply Retweet Označi sa "sviđa mi se"
Blake Camp 31. sij
Odgovor korisniku/ci @NPCollapse
but i take your point about them simply being chemical signals.
Reply Retweet Označi sa "sviđa mi se"
Blake Camp 31. sij
Odgovor korisniku/ci @NPCollapse
Emotions, particularly empathy, likely arose to increase survival odds. There's some work on this but I will have to find it. In competitive social settings its not too much of stretch to suppose that collective emotions improve the survival odds of the population as a whole.
Reply Retweet Označi sa "sviđa mi se"
Blake Camp 31. sij
Odgovor korisniku/ci @NPCollapse
loosely related, but Q for you: can we afford not to give AI emotions? Imagine reasoning with a super-intelligence incapable of compassion. The idea of a constrained objective, e.g. do-no-harm, is prone to all sorts of interpretations, whereas compassion may be more universal
Reply Retweet Označi sa "sviđa mi se"
Blake Camp 31. sij
Odgovor korisniku/ci @jeffclune @reworkAI @reworkdl
Id also love to hear your opinion on whether outer-loop is akin to evolution. Imho it is not, because its a fundamentally different optimization procedure, but the analogy still makes sense. Can the outer loop truly be open-ended in the same way that evolution is?
Reply Retweet Označi sa "sviđa mi se"
Blake Camp 31. sij
Odgovor korisniku/ci @jeffclune @reworkAI @reworkdl
Very interested to see it! So much potential in this direction.
Reply Retweet Označi sa "sviđa mi se"
Blake Camp 30. sij
Odgovor korisniku/ci @jeffclune @reworkAI @reworkdl
Have you actually published the ANML work yet? Currently working on something similar, but not clear about some of your details.
Reply Retweet Označi sa "sviđa mi se"
Blake Camp 30. sij
Disagree, curriculum is a better word for this.
Reply Retweet Označi sa "sviđa mi se"
Blake Camp 30. sij
Odgovor korisniku/ci @hardmaru
True
Reply Retweet Označi sa "sviđa mi se"
Blake Camp 30. sij
Odgovor korisniku/ci @hardmaru
What's ur opinion about writing blog posts as a phd student, potentially sharing your good ideas? Many of us love discussing the best ideas floating around. Then again, we need to retain the best material for publications. Does everyone feel this way?
Reply Retweet Označi sa "sviđa mi se"
Blake Camp 29. sij
The senility of our senior citizens is no laughing matter, and in that regard I sympathize. But, this is utter lunacy. Comical, embarrassing, downright weird.
Reply Retweet Označi sa "sviđa mi se"