wmf

A new team, probably focused on better data.

There have been plenty of open source projects that view themselves as natural monopolies and treat the existence of alternatives as something akin to high treason. They never admit this of course.

It uses Xe cores that come from GPUs.

If you want Nitro you know where to find it... and what it costs.

It's not worth it and I'm kind of baffled that Valve let this happen. https://www.youtube.com/watch?v=BrRIhq2ZdLw

There is a lot of legacy IPSec out there. For example AWS Site-to-Site VPN seems to still be IPSec.

Weird hardware is just worse. You can't beat mainstream CPUs or GPUs that now have decades of investment behind them.

Hard disks aren't even rated for one decade.

Obviously the "give me the answer" style of AI should be banned for students. Do good educational AI tools exist yet?

You keep talking about Gbits but Homa is about latency.

1. You have to use timeouts just like TCP has a SYN timeout.

2. TCP doesn't have encryption either. They should probably show DTLS working though.

4. You can't just say QUIC is faster without testing it.

The packet sizes should be the same for TCP or Homa; the difference is in sender-based vs. receiver-based congestion control.

Flow control is better than nothing but it can cause congestion spreading and bufferbloat. QCN, Falcon, and Ultra Ethernet provide much better congestion control for RoCE but they also require newer hardware compared to Homa.

I've been hearing about Homa for years and wondered what's new. I found a changelog in the readme of the git repo: https://github.com/PlatformLab/HomaModule

As a lone person who owns ten million killbots they could do a lot.

If you don't pitchfork the elites you have no one but yourself to blame.

The real argument about Transport Stream isn't its age but unnecessary overhead. If backwards compatibility is more important than overhead, fine. But acknowledge that there is a tradeoff.

AFAIK the M4 and M6 reverse engineering is AI-assisted.

Discussion from 6 days ago: https://news.ycombinator.com/item?id=49838111

Starshield is pretty cool.

V8 is extremely optimized for script startup time.

"In the past, requests went out to a hosted execution service."

They were outsourcing to another company so there's plenty of room for overhead to creep in.

The point is that kids can't break the OS.

Often they don't and it's kind of the point. You could go by Rich, Richie, Rick, Ricky, Dickie... if you choose to call yourself Dick, you know you're going to cause some amount of snickering and you're doing it on purpose.

Also I remember men called him Richard Feynman but a lot of women called him Dick. That's sending a different message.

Artists and authors genuinely believe their work is being stolen by AI. That part isn't a psyop.

There are a couple of data centers that are causing real water, noise, or pollution problems and it doesn't help that one of them is run by Elon Musk.

The chip shortage and resulting price increases are also very real.

And AI is not profitable. There probably is a bubble that will bankrupt some companies (maybe not the ones you've heard of).

When you mix all these facts with pre-existing skepticism of capital and corporate power you get the backlash. It's exaggerated and spinning out of control, but it's rooted in reality.

I would assume clouds count as resellers not customers.

Not including Q1 and Q2 numbers sounds like a hit piece.

It's called Hyperloom.

If the ROCm team can't get enough compute it doesn't bode well for World Labs.

It has a separate address space separated by PCIe so even escaping the hypervisor won't give access to DPU memory.