GaggiX

49 ms·

931.
by GaggiX 2y ago
>Where do the evaluations come from? Good datasets are selected empirically, they are usually a mix of different sources, not a single engine. >The idea that Stockfish isn't benefiting hugely from Google having created and advert
932.
by GaggiX 2y ago
>It's a neural network trained on self-play games It's not, it's just supervised learning on evaluations. There is no self-play involved when training the model.
933.
by GaggiX 2y ago
>all its ideas were absorbed into Stockfish That's not true at all, Stockfish still uses only human heuristics for search and NNUE for eval, a completely different architecture than alphazero and derived from the Yu Nasu Shogi engin
934.
by GaggiX 2y ago
I'm not complaining, Google can do whatever they want, it's just not proper research, even though they want it to feel like it is.
935.
by GaggiX 2y ago
Because it's presented as research, but the results are not reproducible without at least information about the dataset.
936.
by GaggiX 2y ago
It would be cool to see an airplane made using generative design.
937.
by GaggiX 2y ago
>gives often the wrong intuition when thinking about applying gradient descent on high-dimensional parameter space Can you give some examples?
938.
by GaggiX 2y ago
He meant 2Wh/$.
939.
by GaggiX 2y ago
You should look into server rack batteries: EG4, SOK etc...
940.
by GaggiX 2y ago
I thought about "spatial computing," but I discovered that it was used before the Apple vision.
941.
by GaggiX 2y ago
I'm surprised they call AI AI and not some made up name like Digital Crown. EDIT: better examples, Taptic Engine, Touch ID.
942.
by GaggiX 2y ago
Why does Microsoft spend resources on their own model when they can freely use OpenAI models? The model must have some specific characteristics.
943.
by GaggiX 2y ago
Have you seen the latest VAC ban wave?
944.
by GaggiX 2y ago
The new VAC ban wave has been achieved using machine learning on the server side.
945.
by GaggiX 3y ago
>For a given maximum number range, it's trivial to make Miller-Rabin actually deterministic. You just choose bases that have been proven to together exclude all pseudoprimes in the given range. What are the bases for the range of 10
946.
by GaggiX 3y ago
I believe that most people with a plug in hydrid actually forget to plug the car in, so the car is almost never in electric.
947.
by GaggiX 3y ago
There was a repo that falsely claimed to have released the language action model code, when it became apparent that the "leaked" code did not actually prove any of the points made in the README the author deleted the repo. You can
948.
A Careful Examination of LLM Performance on Grade School Arithmetic(arxiv.org)
2 points by GaggiX 3y ago|0 comments
949.
by GaggiX 3y ago
There are many groups in the US that live with very little use of advanced technology, but very few people have joined them from outside, the vast majority leave after experiencing the lifestyle for some time, modern conveniences are hard t
950.
by GaggiX 3y ago
Third paragraph: "We believe that the same way mobile payments leapfrogged credit cards in some markets, and mobile phones leapfrogged desktop computers in developing economies, AI too will (at least initially) leapfrog more legacy tec
951.
by GaggiX 3y ago
You should have read the article, they explained it in the second paragraph, and I have no idea why anyone would assume it's because of AI.
952.
by GaggiX 3y ago
>I don't want AI. Don't use it. >I want to build my own house, grow my own food, make my own shoes. AI is not stopping you from doing these things.
953.
by GaggiX 3y ago
>Another assumption is that it’s because of tokenisation issues. But that can’t be true either. It's definitely a tokenizer issue, if GPT-4 was trained on singular characters I'm pretty sure it would be able to play Wordle much
954.
by GaggiX 3y ago
With Llama 7 8B, Groq still has high API limits, so you cannot use it at scale for a project.
955.
by GaggiX 3y ago
The nice thing is that you can now use Llama 3 70B for very cheap ($0.9/M tokens) using FireworksAI, TogetherAI or similar providers.
956.
by GaggiX 3y ago
Deleted by the author probably because of: https://web.archive.org/web/20240424095544/https://github.co...
957.
by GaggiX 3y ago
Get your OpenAI API key and then use it on one of the hundreds of open source frontends available, such as: https://github.com/lobehub/lobe-chat
958.
by GaggiX 3y ago
>Have to pay $20 just to try it out. No you don't, API are available to the public and there are plenty of frontends, the API are convenient even you use GPT-4 often. Prompting GPT-4 would have cost 0.01$.
959.
by GaggiX 3y ago
I'm talking about translating English into Georgian for Georgian speakers, thanks to neural networks there are less reasons to learn the dominant international language as such protecting smaller languages.
960.
by GaggiX 3y ago
Claude 3 (even Haiku) can speak Georgian perfectly, if more than half of the websites are only in English now you can translate them very well into hundreds of different languages thanks to neural networks, globalization is the reason why t

More ›