cayleyh

And also just the flat hardware failure rate.

Not a lot of public data about things like that at scale, but there are some things like the Meta Llama report that had ~ 9% annualized gpu failure rate while training Llama 3. Not 100% clear from the Meta report if those are dead hardware or just 'unhappy' because they just put them into the 'interruption caused by' bucket.

Would be interested if anyone had better data on actual live failure rates of h100 or better class gpu that suck so much power.

https://ai.meta.com/research/publications/the-llama-3-herd-o... https://www.trendforce.com/news/2024/10/31/news-datacenter-g...

*staring blankly, mumbling "leave Selena alone" over and over and over again as I walk out into the ocean*

Specifically I think this will be true around defence procurement: Canada was just added as the first non-EU state in the EU SAFE defence agreements, and just did a major booth at the International Defence Industry Exhibition EU defence trade show. They decided to buy new subs from Germany instead of Korea, and have slow-played the new fighter acquisitions and will potentially replace at least some F35 orders with Saab Gripens. And they have a new deal to start manufacturing Ukrainian designed military drones in Canada (partnership with Ukraine announced last week iirc?).

If getting this status helps simplify process and costs of EU<->Canada military procurement, then it will be a clear win for both sides given the current desire to reduce dependence on US defence providers.

It could be, but common, we all know that Anthropic's watermark is the using "load bearing", "genuine", and "seam" 1000x more in the same paragraph than any human in history.

Better yet, he gets loans, collateralized against those unrealized gains to the current value of the gains, and then writes off the loan interest against his present and future taxes! Amazing!

DDR5 is DUV not EUV

the civil war was only ever and exclusively about states rights

Dario Amodei in David Attenborough voice: "This Claude appears to think more frequently and more deeply to give better responses"

I guarantee the "median" payout is nowhere near that. The article said the average pay is 158 million won. I bet the median pay is less than half that.

> I personally use my OpenAI subscription pretty heavily, 2-3 agents running practically all day on various tasks but I never even get close to running into limits

Same. But if I was working for an organization that measured token usage, you can bet I would be doing things like creating a cron job that uses claude to create a customized bespoke report update of the current status of all my open assigned tickets and message that to myself 4 times a day... token burn for zero purpose whatsoever.

seL4 + linux kernel api shims does seem like the most realistic approach to move forward. But the driver + driver API situation seems like the biggest barrier.

A compounding problem is that lots of the key drivers depend on binary blobs provided by the vendors, making updating and porting that much harder. I know there has been some work to have translation layers between "linux driver api surface" --> "other OSs" but that feels problematic to trust & maintain unless it's driven from the linux kernel team itself.

I missed what the reason given for moving from zig to rust was. Considering 9 days ago it was "This whole thread is an overreaction."

Is it really just that Anthropic now owns bun and zig is "anti-LLM" for contributions? Was there more justification given elsewhere?

The PR says it was from Jarred, and then he flags his own PR as slop...

100% agree on this. Even just the flashing on nav menus. Very very common to load an app I'm already signed in to, see the "login / signup" state first, then a loading indicator or skeleton, then the logged in state. Just seems like amateur hour all over the place of people or companies not caring.

You can buy them frozen, directly from the baking counter, in 144 count boxes... Just saying...

The ram is "unified" meaning it's a single shared between CPU and GPU, and it's "on package", meaning the RAM chips are packaged together with the CPU / GPU die, but it's just regular old RAM chips.

You can clearly see this in the shot of the Mini mobo: CPU/GPU ASIC with 2 separate ram chips packaged next to them: https://cdn.wccftech.com/wp-content/uploads/2024/11/Mac-mini...

We're all living in a William Gibson novel now...

I too feel this is a huge part of it, coupled with the fact that "basics" of last generation (a home you own, a stable job that doesn't overwork you on evenings and weekends, affordable options to have a family) are also being priced out of many peoples lives. You feel like you're not matching what your parents and cultural artifacts tell you you should be achieving at your age, and at the same time you're flooded with influencers on ski trips to Japan or snorkelling in Jamaica every other weekend, and it's a perfect recipe for feeling bad about your life no matter how well off you're doing compared to yesterdays median statistic.

It's hard to get a comparable read since you don't get a clean split in the #s between what would be public spending vs. private spending if the US + a "Medicare for All" type system, but including the % of GDP spent in US on healthcare overall, it would put government expenditures as % of GDP on par with most other countries in the world that do provide universal health care:

* https://www.imf.org/external/datamapper/exp@FPP/USA/FRA/JPN/... * https://www.healthsystemtracker.org/chart-collection/health-...

France certainly has a higher % of expenditure to GDP than other comparable countries, and you would expect the USA health care to GDP % to decline to be more inline with other countries with universal coverage if a national program was introduced.

However, because France is still offering more public social services and benefits overall vs. a "USA + universal health" that it's hard to make broad claims either way about who is wasting more money or which system is more effective for citizens based purely on % of government expenditure to total GDP.

What does your setup look like? What kinds of workloads are you using it for?

Underrated acquisition. Gives NVIDIA a whole lineup of inference-focused hardware that iirc can retrofit into existing air cooled data centres without needing cooling upgrades. Great hedge against the lower-end $$$-per-watt and watt-per-token competition that has been focused purely at inference.

DTM is my new preferred power generation measurement unit :D thank you!

Also AWS Graviton and Google Axion servers & VMs on those clouds

"decade" being the universal time frame for "I don't know" :D

"because of the monolithic nature of ZFS that created problems with the Redox microkernel design"

Anyone have an idea what this actually means and what problems they were having in practice?

"Half of all developers have below-average diligence" - a lot of this is also not developer choice, but environmental. So much software is developed and maintained in very constrained economic environments, often by solo devs who also have other responsibilities. The choice here often is trading some "diligence" for "meeting business requirements in the time / budget constraints" imposed by your employer.

The company behind it got bought by Oracle, so all the public stuff was archived. Not sure if there is still any community around it after that.

I structure all my personal relationships around B2B SaaS contracts. Mom isn't happy when I make changes to the ToS but I've got pretty hard vendor lock-in and a defensible moat there, so I mostly don't worry about the reputational damage.