Hacker Newsnew | past | comments | ask | show | jobs | submit | evolveyourmind's commentslogin

Have you successfully submitted a “malicious” paper that hacks these AI reviewers?



That's a great concise implementation! I'm glad you also went with the parse-and-interpret-an-AST approach: it seems like the right design. Your `Reconstruct` type is a great idea! I might copy that.


I can't understand why they invest so much in gaming. Gaming on such small devices is too uncomfortable, both for your eyes and your posture. In addition, with cloud gaming you can basically run any PC/Console game on any device provided a good-enough connection (eg. Nvidia NOW, Xbox cloud gaming etc)


Mobile gaming is about 10x the size of the PC and console markets.


And 90% of money spent on mobile is for some sort of virtual item slot machine.


But how much % of these mobile gamers are using iPhones, I wonder.

AFAIK developing countries are dominating mobile gaming for a simple reason that they cannot afford PS/Xbone/PC. Ergo no iPhone for them to begin with.


Interesting number, but how many are actually that GPU-intensive?


A surprising amount. Genshin and many Gatcha games are GPU intensive, even on such a small screen.

It might be because they code very inefficiently, and they might also be doing a ton of crypto and anti-hacking calculations in background, but these games are surprisingly taxing for the phone they're running on.


I think a substantial amount at least. Kids and tweens seem to game a lot on mobile, and Fortnite is a GPU-intensive game that runs on mobile.


It's literally not on iOS though


A very large portion of Apple's ever-increasing "services" revenue comes from mobile gaming (30% of IAP).


Because gaming is one of the last remaining markets that apple can dominate in and profit. Microsoft spent a lot of money on Activision to claw back against apple

If your eyes hurt then switch to literally any other apple device, Apple TV, Mac, iPad, iPhone… they all have access to Apple Arcade. When cloud gaming streams asset objects and not raw video, then it’d be a contender. As of now, it’s too expensive and is losing to mobile games

https://imgur.com/a/BG2hexr


MS is buying Activision Blizzard to compete against Sony.


The other comments are right in saying mobile gaming is huge.

But, if you are referring specifically to the new GPU with accelerated raytracing, I want to add something more.. Remember that for many years now, every innovation that Apple has produced has been secretly developing towards AR!

For this reason it makes perfect sense.. Raytracing is SUPER important for the kind of rendering needed to make AR/VR look realistic


You have no idea how much the Chinese spend on gaming. They're basically doing a Blizzard. Don't you guys have iPhones?


You can also easily implement meta-languages. Eg. lambda calculus: https://github.com/desi-ivanov/ts-lambda-calc


Any benchmarks?


Some people have timed it here, it looks like it's taking 15-20s/it (dependent on quant and hardware).

https://github.com/leejet/stable-diffusion.cpp/issues/1


I have compiled it with the command:

cmake .. -DGGML_CUBLAS=ON -DCMAKE_CUDA_COMPILER=/opt/cuda/bin/nvcc

to use my NVIDIA GeForce RTX 2060 SUPER.

I have converted the model to use FP16.

With these choices, the time per iteration is between 8.5 s and 9 s and the total time for making an image is around 200 s.


That seems a lot worse than a 2060 SUPER with PyTorch in A1111.

https://vladmandic.github.io/sd-extension-system-info/pages/... (search for 2060 SUPER)


This is not surprising, because the GPU support in GGML is said to be preliminary and it is optimized for being run on CPUs.

Seeing the times reported by other people, it seems that using the GPU with GGML, instead of the CPU, still provides a speed improvement, but it is small.

Nevertheless, I have appreciated that after following exactly the instructions of this project everything was up and running after a few minutes and it could be tested.

Past attempts to install all the environment needed to run such models have required much more work.


Other than Monads, HKT can be used to easily write type-level functional programs [1]. This can for example help writing type-level parsers for other lanugages.

A real world use-case could be parsing GraphQL raw string queries and automatically infer the returned types based on a common schema, without using special code-generators. For instance you can come up with some magic function `gql_parsed` like:

doc = gql_parsed`query GetUser { user { name }}`

where doc is inferred as something like Doc<Query<{GetUser:{user:{name:string}}}>>

[1] https://desislav.dev/blog/tsfp/


Seems very limited. I wonder if the same can be achieved with just stable diffusion and neighbor latent walks with very small steps. On the other hand the interpolation techniques with the GigaGAN txt2img produce much higher quality “videos” than this


Right, stable diffusion and control net depth network could give very good results if you have a source video.


Imagine the impact of such a crash to the whole humanity in the next years. Api crashes and suddenly all bots crash and half of the world is stuck for some hours


“We have to push back the product launch date. The AI bot is down and that one employee that knows how computers work was laid off due to redundancy.”


Luckily it currently seems that AI will be federated. No actor has a proper moat for even short-term competition to catch up.


A containerized version of this thing would be def useful, as it installs global packages and assumes a lot of preinstalled binaries. The node image won't work alone tho, you'll python, pip, git, cpp compiler


Yeah, I've been wanting containers for these type of projects for a while now. Conda is fine if you're already involved in the ML/Python ecosystem, and as an outsider to that world I guess I have no right to complain (Conda is actually not all that hard to learn all things considered), but boy would it be nice if I could just install Docker, run `docker run cool_project/ml_wizardry`, and have a demo up and running in my web browser instantly.


Would nix be a good fit for this?


Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: