redox99

40 ms·

871.
by redox99 3y ago
*Prompt:* I'm a software engineer who has never done anything with AI/ML. Explain to me transformers and the "Attention is all you need" paper. *GPT4:* "Attention is All You Need" is a seminal paper published i
872.
by redox99 3y ago
The trend has been going from domain specific models to things like ChatGPT.
873.
by redox99 3y ago
I love how every time this EU thing is mentioned, you already know beforehand the 5 companies they'll mention as a "counterpoint"
874.
by redox99 3y ago
People are extensively using these models (more specifically the finetuned, uncensored ones) for role playing.
875.
by redox99 3y ago
Except the part about "with some changes that make it more difficult" is false. The changes make it trivial. Edit: Omitting that part still makes GPT4 over complicate the solution
876.
by redox99 3y ago
Is this a custom model, or based on OpenAI's API?
877.
by redox99 3y ago
More than half of that is India. And high end vehicles such as the Model S are probably very rare there. So although true that it's big enough to sustain right-hand-drive car production, it might not be big enough to sustain high-end c
878.
by redox99 3y ago
Cybertruck is mostly a truck for the US market, it won't sell much in the UK. The "model 2" however will definitely sell very well, probably outsell the 3 and Y combined in the UK (as the average car bought in the UK is much
879.
by redox99 3y ago
I'd argue that books are a clear example where the 100k tokens context would make a huge difference.
880.
by redox99 3y ago
Tell it the columns of a certain table. Ask it a GraphQL schema for a CRUD for this table. Ask it for a resolver using X libraries. Ask it for unit tests using Y libraries. Couple hours work done in a few minutes.
881.
by redox99 3y ago
I guess if you do that you're not too far from planar reflections in terms of cost, which would be better as well since they don't suffer from occlusion artifacts.
882.
by redox99 3y ago
By that logic, Doom must be smarter, otherwise it wouldn't run on significantly weaker hardware.
883.
by redox99 3y ago
There's nothing brute force about getting this level of quality with just 2 samples per pixel. It is pretty magical actually, only possible thanks to ReSTIR, AI denoising, AI upsampling, AI frame generation and strong hardware accelera
884.
by redox99 3y ago
4090 with DLSS upscaling + framegen is almost there, so next gen for sure.
885.
by redox99 3y ago
For something like CP77 with path tracing: yes. The extra horsepower + DLSS frame generation will give you around double the FPS.
886.
by redox99 3y ago
Most games (including Cyberpunk without the path tracing mode) do use hybrid renderers indeed (so it's sprinkling ray tracing to specific parts of the traditional pipeline). The path tracing mode mentioned here however, does "full
887.
by redox99 3y ago
To be fair, the fact that I worked in game development for a while makes the typical rasterization inaccuracies stick out like a sore thumb to me. Kind of like once you see it, you can't unsee it. And there's also a bit of knowing
888.
by redox99 3y ago
Playing Cyberpunk Path Tracing on a friend's 4090 was incredible. Over 100fps (with all the upscaling/framegen trickery). The visual jump is significant enough that I find it worth to power through the 40-60fps I get on my 3090 (a
889.
by redox99 3y ago
> Why do so many commenters on HN always sound so confident, yet none of them work in publishing? It's pretty simple. Most of us here are used to just downloading "preprints" pdfs from arxiv for basically any CS/AI pa
890.
by redox99 3y ago
Proven in what way? As a way to lose money? It's no surprise that it's easier to get to Level 4 if you have a 50-100k sensor suite, HD maps and a geofence. However, I wouldn't be shocked if, even if Tesla robotaxis arrive man
891.
by redox99 3y ago
> I'd wager you don't want fighter jet or missile guidance software to be open source. I'd be OK with that.
892.
by redox99 4y ago
Because it's over 9000
893.
by redox99 4y ago
GPT4 32k + some tricks probably fixes that.
894.
by redox99 4y ago
It's pretty simple actually. Get a 3090 or 4090. Forget about AMD.
895.
by redox99 4y ago
Yes, I also expected GPT4 to get the joke, as I've seen it understand similar jokes.
896.
by redox99 4y ago
(GPT4) In this script from a Doctor Who episode, Clara and the Doctor are having a conversation about the internet. Doctor Who is a British science fiction television series that follows the adventures of the Doctor, a Time Lord from the pl
897.
by redox99 4y ago
Here's a couple more from GPT4 (since it's random every time because of temperature) GPT-4を翻訳に実際に使ったことがありますか？本気で、伝えたい意味だけを伝えるという話は、ちょっと試してみれば簡単に解決できると思うのですが。実際にGPT-4を翻訳に使ったことがありますか？本当に、試してみるだけで簡単に払拭できると思うのに、この「明確な意味だけが伝わる」話ばかりで。
898.
by redox99 4y ago
GPT4 実際にGPT-4を翻訳に使ったことがありますか？本当に、明示的な意味だけを伝えるという話が、試してみるだけで簡単に解決できるなんて、冗談じゃないですか。
899.
by redox99 4y ago
There is a very clear benchmark here though. Can you use GPT4 translations "as is" for your product[1]? I've used GPT a fair bit for English<->Spanish (both directions) for my product, and I often have to fix the odd wo
900.
by redox99 4y ago
It's interesting that there's a lot of people that will not buy a Tesla because of Elon Musk's politics/twitter/whatever, but at the same time they have no problem buying a VW, after Dieselgate (not saying that&#x27

More ›