redox99

48 ms·

961.
by redox99 4y ago
What surprises me is that it goes beyond not releasing weights or not releasing source code. They haven't even published a paper on GPT 3.5, ChatGPT, ChatGPT turbo, etc. They haven't even said what ChatGPT turbo is, beyond "i
962.
by redox99 4y ago
I'm pretty sure if let's say OpenAI worked 1 year on passing the Turing test, they would achieve it. It seems extremely close. Current LLMs are trained to be useful, not to trick humans.
963.
by redox99 4y ago
Your info is out of date. The 4680 actually ended up being 244Wh/kg[1], which is lower than the Panasonic 2170 at 269 Wh/kg that they were already using. [1] https://www.youtube.com/watch?v=4XOHetABrag
964.
by redox99 4y ago
Coinbase is relatively trustworthy (publicly traded US company). I believe trading stablecoins has 0.001% fee, and ACH withdrawals and deposits are free[1]. And on the crypto side, USDT transfers on Ethereum (the most expensive chain) are u
965.
by redox99 4y ago
Probably a very unpopular opinion here in HN, but stablecoins (USDC, USDT) are among the best ways to send money internationally. It's so much better than SWIFT.
966.
by redox99 4y ago
> To read means to understand and comprehend for me. Reading and understanding are obviously different things. I can read a paper on quantum physics, but I can't understand it.
967.
by redox99 4y ago
It's just an example to make the point easier to understand A single engineer with [some near future LLM] could potentially do the work of [X] engineers right now.
968.
by redox99 4y ago
You can literally finetune these OpenAI models using their API. In this case it probably wasn't done because the author found it too much work and/or too expensive.
969.
by redox99 4y ago
AGIs should basically by definition be able to replace humans[1], so yes, when we get good enough AGIs that will happen. GPT4 won't be an AGI of course, but it still could have a massive impact on the job market. A single engineer with
970.
by redox99 4y ago
> Bottom of the heap? Ai essays suck. They’re painful to read for the most part. I'm just responding to that part
971.
by redox99 4y ago
ChatGPT is way above your average high school level.
972.
by redox99 4y ago
It's ancient hardware, and FUSE is very slow. A benchmark on modern hardware would be interesting.
973.
by redox99 4y ago
You can always mount an FTP server and put your swapfile there.
974.
by redox99 4y ago
Basically ENS (Ethereum Name System) but simpler and cheaper.
975.
by redox99 4y ago
$12 is too high if it's not going to remove ads.
976.
by redox99 4y ago
Maybe they changed their mind on that, because it does not show me any DRM usage as of now.
977.
by redox99 4y ago
I don't believe DRM fingerprinting is used in the wild. Firefox shows when DRM is being used (like Netflix) and I've never seen it used outside that.
978.
by redox99 4y ago
EDIT: The title has been fixed now. The title was incorrectly altered. Where does the "14th Tesla with FSD emergency vehicle crash" in the title come from? The article never says this vehicle was using FSD. It says it might have b
979.
by redox99 4y ago
I assume they mean instead of minimax (ie assume perfect from the opponent), to play around the flaws of the opponent. If you know the opponent won't spot a certain line, or will choose a certain sub optimal line, you would have an adv
980.
by redox99 4y ago
Has there been significant progress in Go in terms of AI strength after AlphaZero (like there has been with chess), or has Go AI stagnated because of lack of interest and/or resources?
981.
by redox99 4y ago
> True, but this entire loop happens within the model No, it's literally a for loop in python that runs the whole thing from scratch[1] after appending each new token. No artificial slowdowns, what you're seeing is literally wh
982.
by redox99 4y ago
Its input is a sequence of tokens (so text), and its output a list of probabilities for the single next token. You pick the highest probability token, append it to input and execute the neural network again. If it outputs an "<end
983.
by redox99 4y ago
Yes it would, or at least a similar operation. The point of using a CNN instead of a FCN is that you force it to learn in a certain way that prevents overfitting. But given a sufficient dataset, and proper data augmentation you would expect
984.
by redox99 4y ago
Really happy for you. I thoroughly enjoyed your YouTube NN series, and your Lex interview.
985.
by redox99 4y ago
The causal implication in the title is basically made up (8% drop entirely because of incorrect example). The way I see it, in this AI war, considering google already has an almost monopoly, they have a lot to lose and not much to win, an
986.
by redox99 4y ago
I tried (regular) bing after this announcement, and I'm kind of surprised at some of it's flaws. 1. Visited and unvisited links look the same???? (Instead of blue vs purple). Literally a deal breaker in my case 2. No dark mode in
987.
by redox99 4y ago
Considering ChatGPT reached 100M users in 2 months, it's clear a lot of people care.
988.
by redox99 4y ago
Source?
989.
by redox99 4y ago
GPT3 very own paper, "Language Models are Few Shot Learners"[1] show otherwise [1] https://arxiv.org/abs/2005.14165
990.
by redox99 4y ago
I've actually used it quite a bit and he's not "flat out incorrect". I'd say you're the one wrong. It might reply that when you ask it something that it can deduce it can't know, but ask it something reaso

More ›