russianGuy83829

Thank you, looking forward to it.

I just saw this simple patch to enable MTP (potentially 2x performance) on older GPUs (Kepler etc), so maybe it will work for you

https://github.com/ggml-org/llama.cpp/pull/25680

Also, for Qwen, the 4 bit _XL quantization seems to have a good balance of performance to size.

That's great. Personally, I'd interested in Qwen3.6-27B and deepseek V4 flash (or pro), with contexts above 60k. They seem to be popular and have good coding performance. I'd appreciate numbers on a single or two GPUs where a quantized version fits reasonably into the VRAM (Qwen in 16 or 24GB). 4 older GPUs approach a used 3090 in price, and the 3090 has better support for speedups like MTP. So cheaper but slower looks like a reasonable target to me.

Have you tried 27B class models like qwen3.6?

you can close the dialog and redo it after answering wrong. the correct answer is usually the longest or the most specific. the latter half of the questions doesn't take the setting into account, it's reinventing technologies that would already be there some summaries on the main page (steel) leak answers

Not sure if its intel specific, but for the amd variants you can download an .iso instead of an .exe and boot from that to upgrade. No need for windows

CNN; Midterms

bing (use desktop mode) has both claims https://www.bing.com/search?q=https%3A%2F%2Fservury.com%2Fda...

archived versions here: http://archive.today/8LX8s http://archive.today/00mIw

Yes, the page mentioned ISO27001 which is still visible in the indexed duckduckgo result.

https://duckduckgo.com/?q=https%3A%2F%2Fservury.com%2Fdatace...

It is not visible in the live webpage.

Is it april fools yet? Whats next, tumblr merging with boston dynamics? The stock is up 30% and I bet there has been insider trading right before the announcement

previous pixel phones also had custom Google silicon, just with some Samsung IP

like every browser identifying as Mozilla in the user agent

GrapheneOS is seemingly working with an OEM to make a GrapheneOS smartphone. Its probably not samsung, but would still be an established vendor

The auto-sending part seems to be speculation: https://www.resetera.com/threads/gaming-copilot-installing-a...

I mapped it to forwards and back buttons, it works well

Found it better than the MX Master, especially with extra keys remapping and more durable materials.

<Furiously typing GDPR data request..>

You don't have to weaken your encryption to your bank though. The author proposes extending the TLS protocol so that the bank declares themselves responsible for the contents of communication and full strength encryption can be used.

@dang all users here are probably bots

I don't see the mental gymnastics here, the post you linked pretty clearly delineates the differences to AI in the sense it is used today.

As major differences I'd highlight: local and offline, so drawings not sent anywhere trained on artist work with explicit consent

In Krita's case, they claim the AI isn't generative so it doesn't add detail.

Whereas the AI today is trained on stolen work and often on the inputs as well.

I see it fine when zooming in a bit on mobile

that's going to be a banger bobbybroccoli video

any open source alternatives?

Is there a repository to reproduce this?

Do you happen to have wall power consumption readings at idle for your Mini-PCs?

people live on the island with the Statue of Liberty: https://nypost.com/2019/10/12/what-its-like-to-grow-up-next-...

good luck to you on your endeavors! I’m sure the HN crowd will be interested in whatever comes out of this ツ

sorry about the off topic, but has anything come out of this? https://news.ycombinator.com/item?id=39142069

It seems like this can’t run all models, and needs custom ones trained from scratch: “ We introduce two new models: TurboSparse-Mistral-7B and TurboSparse-Mixtral-47B. These models are sparsified versions of Mistral and Mixtral […]. Notbly, our models are trained with just 150B tokens within just 0.1M dollars”.

It remains to be seen how good these custom models are.

Finally sideloading and real firefox on iPads! Hopefully virtualization and JIT will follow, making it a bit closer to a real laptop replacement.

This doesn't address the coincidences that the GRU unit is consistently at the place where the "headaches" happen, get identified by the victims, and work with one of the only people (Russian professor) researching the rare disease that was diagnosed for one of the victims. There are documents in the article that link the group with the researcher, which explicitly mentions acoustic non-lethal attacks.