wmf

Some people do run Tailscale proxies in their home, and an always-on device like a TV is a good place to run it.

Usually silent data corruption isn't caught. If a file fails to open you might realize it's been corrupted.

Never? They're currently renting out those GPUs at a profit.

Anything with RAM in it is an appreciating asset.

H100s (and the PCIe converter card you need) are available on eBay.

The list price of cloud is 10x higher than the price of hardware itself so that leaves room for some discounting.

I think Xiaomi already distilled the Porsche Taycan.

AFAIK you can't really print crystalline silicon so they have to start with a wafer and dope/etch into it to form transistors.

UPSes are designed to run for minutes not hours. I guess enough Megapacks to run overnight would be more expensive than gas generators.

Most open models are developed for profit so they should be equally liable.

For whatever reason, nobody is trying to build solar-powered AI datacenters. Maybe they can't find enough land.

Ironically, Flock might create enough backlash to trigger regulations against surveillance. They're certainly an example of what not to do.

"The other “local AI” apps you’ve heard of? They’re proprietary shells built on top of open-source engines they don’t own."

This is a roundabout way of addressing LM Studio.

Frontier open models are Kimi K3, GLM 5.2, DeepSeek V4 Pro, etc. They're all too big to fit on most Macs.

Somehow a certain set of labs are all releasing open weights and a certain other set of labs are closed weights.

AFAIK it's been documented by Los Alamos since they're at high altitude.

It's a Vulkan extension to use dedicated video decoders like QuickSync, NVDEC and VCN. It's not using shaders.

ECC UDIMMs are crazy expensive, presumably because they don't have economy of scale. Mandating ECC would solve that.

"This, however, is absolute garbage and nonsense. The last thing you want is more regulations and government certified systems." — the guy who designed the Ford Pinto

IIRC Itanium was faster on natively compiled HPC code, presumably because it had more FPUs and larger cache than the Pentium 4. But in general it probably wasn't a good upgrade and Intel would have had to sandbag x86 a lot to force Itanium adoption.

Itanium 1 had hardware x86 compatibility and Itanium 2 had software x86 emulation just like Apple. It wasn't enough.

The patent pool system is really bad. There's no incentive to weed out bad patents.

If you want a particular outcome, legislate that outcome.

Every photo will be AI-enhanced and correctly labeled as such.

Just like every Web site has a cookie warning.

That kind of regulation has failed over and over. The obvious outcome is that every listing will have misleading AI "photos".

No one cares.

The community probably would have complained if they registered the trademark preemptively. But if they let the bad guys register it and then save the day it's the least bad solution.

No? Decompression is rated in GB/s per core now.

High voltage DC over 50 V is usually considered pretty dangerous.

Do you think people who work on AI for a living are not going to use it?