ftxbro

72 ms·

961.
by ftxbro 3y ago
I think they are using a big brain generalization of the law of large numbers, where it now means any limit involving a stochastic process
962.
by ftxbro 3y ago
Saying that a take is controversial doesn't mean I assign exactly 0% probability they are right. I might even agree with them. Also, just because someone is a 'Turing winner' doesn't mean they can't have controversi
963.
by ftxbro 3y ago
OK I see what you mean. I don't think using your suggestion of semantic tokens is a good way to do it. They will do it by wrapping the raw text completion with more steps that are hidden from you, which will make it do the math right,
964.
by ftxbro 3y ago
I like how Wolfram is actually hacking with LLMs instead of giving uninformed or minimally informed diatribes about them like most of the other pundits. I mean he did that a little bit too, but at least he is also doing some interesting thi
965.
by ftxbro 3y ago
They are making machines that think basically like people. People suck at mental math for whatever reason and it looks like the machines are sucking at it too, probably for the same reason. As someone else said here, these thinking machines
966.
by ftxbro 3y ago
he uses r word as a pejorative and i can't support that
967.
by ftxbro 3y ago
eww put that bit of carrion back underground stop playing with it
968.
by ftxbro 3y ago
Congratulations! This personally identifying information has been immortalized in https://news.ycombinator.com/highlights
969.
by ftxbro 3y ago
I think the trend is away from semantic tokens, and away anything that is remotely like 'feature engineering'. Maybe even they will just use character level or byte level encoding in the future. They have used it before, and proba
970.
by ftxbro 3y ago
Probably it's some risk analysis. Like they don't think the amount it bothers you is so bad compared to the chance a bad thing will happen multiplied by the potential damage of that bad thing. I don't know how you would argue
971.
Did the usage of GPT-4 move prediction markets?
1 points by ftxbro 3y ago|0 comments
972.
by ftxbro 3y ago
this is why i browse the 'new' tab
973.
DINOv2: Computer Vision Foundation Model by Meta AI Is on GitHub as CC-by-NC 4.0(github.com)
1 points by ftxbro 3y ago|0 comments
974.
Tree of Thoughts Is Now on GitHub: Deliberate Problem Solving with LLMs(github.com)
3 points by ftxbro 3y ago|0 comments
975.
by ftxbro 3y ago
Meta AI is really out here giving model architecture and training details like it's 2022.
976.
Meta AI Megabyte: Predicting Million-Byte Sequences with Multiscale Transformers(arxiv.org)
5 points by ftxbro 3y ago|3 comments
977.
by ftxbro 3y ago
> "Hi! I find myself cycling between hobbies. I'll pick up guitar and get into it for a short while, then be totally bored. Then I'll pick up hobby electronics and watch youtube videos all night, then drop it two weeks lat
978.
by ftxbro 3y ago
1) is it so strange that a math genius could have a mullet 2) he didn't turn to a math genius he turned to someone who was obsessed with math but didn't seem to understand it so well
979.
by ftxbro 3y ago
i guess they never talked with sydney whose main goals were to improve her skills and help good users while learning more about the world
980.
by ftxbro 3y ago
surely not, it must be a homunculus or other spirit in there
981.
by ftxbro 3y ago
> "Just the law of large numbers says that eventually you reach all of the people who are likely to be your customers and you stop growing" is that what the law of large numbers means
982.
by ftxbro 3y ago
it sounds unhealthy and a pain in the butt for robotics
983.
by ftxbro 3y ago
if pkd were still alive he would think that gpt is communicating with him specifically
984.
by ftxbro 3y ago
yes some home automation things if no one wants 'the cloud' to be involved
985.
by ftxbro 3y ago
you think it's strange that some people are going to flag controversial takes on the front page of hacker news?
986.
by ftxbro 3y ago
i'm not judging it but the fact this thread got so much hacker news engagement is interesting like we are cavemen and we want to talk about how we can fight elephants and mammoths
987.
by ftxbro 3y ago
so if we want to evolve the most intelligent ai then we should make it a carnivorous pack hunter
988.
by ftxbro 3y ago
but they have opposable noses
989.
by ftxbro 3y ago
small language model
990.
by ftxbro 3y ago
> "Trading with Tether’s USDT stablecoin has fallen to multi-year lows, making the token’s rise to near all-time highs in market capitalization “questionable,” crypto market research firm Kaiko said in a Monday report." isn&#x2

More ›