dvogel

> Annual carbon emissions aren't even trending downward, much less getting to net negative!

They were trending downward in the US. AI usage in the US did change that single-handedly. I understand carbon emissions are a global problem but since they are subject to political remedies, the solution to the global problem is solving the same problem within each political jurisdiction.

Oxide doesn't have the same type of capex opportunities that Amazon had. Amazon had a massive physical footprint to build. Oxide contracts with other companies to produce the actual boards. Most of the inventory of components they have are held in their contractor facilities. So they don't have extra warehouses to build. They could change course and adopt more in-house production. However, I think doing that during the sale of their first model would be unwise.

Isn't the goal of the author to get reproducible behavior out of the agent though? I would thinking turning the temperature down would serve that production goal too.

For these tests, why not tune the temperature and such to reduce the randomness and convert them to almost-always-succeeds vs almost-always-fails? Is it not the iteration count that drives up the cost?

I couldn't agree more. Thus far I'm still objectively more productive than all of the AI enthusiasts I've worked with. I think a lot of the activity with these tools is coming from people who just enjoy using them more than they enjoyed coding. They feel more productive not because they are producing more but because they are producing somewhat less with much effort. It takes them roughly the same amount of time even if it changes the distribution of time spent on each task.

> and in recent weeks it has started to dominate the Twitter discourse.

As a general rule, I don't waste my time with the advice of people who still think Twitter is a source of wisdom.

You're mistaking a compromised build pipeline versus a compromised source repo that only triggers in some build pipelines. You can do reproducible builds from compromised source tarballs. Nothing about reproducible builds necessarily requires source control. Yes, if some people who built from source control compared their builds to the builds from the tarballs it could detect the xzutils compromise. However I have yet to see a reproducible build project that includes such cross-build checks.

No, it wouldn't. The xzutils attacker compromised the source repository. The build pipeline portions were used to obscure the purpose of the exploit embedded in the source code repository.

(Not OP, but...) I still fail to see the current value in confirming that a reproducing builder also included the same compromised dependency that I did when I built it. I understand that reproducible builds are guarding against dynamic attacks within build infrastructure. However I just don't see those happening. Compromised source dependencies are a 100x more common problem.

> US-centric here: I feel that uploading a small percentage of a file as a condition of downloading the whole thing may very well fall under fair use

This is not consistent with current Fair Use application. The TL;DR is that the use must be for a new expression that is protected by either the 1st amendment or copyright itself. I don't know of a single case where it has applied to mere distribution. I would be astounded if there is such a case because that isn't within the expressed purpose of the doctrine.

> But as a standard library abstraction, it’s too opinionated. It categorically excludes cases where sources form a tree: a validation error with multiple field failures, a timeout with partial results. These scenarios exist, and the standard trait offers no way to represent them.

This seems akin to complaining that the CPU core has only one instruction pointer. There is nothing preventing a struct implementing `Error` from aggregating other errors (such as validation results) and still exposing them via the `Error` trait. The fact of the matter is that the call stack is linear, so the interior node in the tree the author wants still needs to provide the aggregate error reporting that reflects the call stack that was lost with the various returns. Nothing about that error type implementing `Error` prevents it from also implementing another error reporting trait that reflects the aggregate errors in all of the underlying richness with which they were collected.

I lived in DC for 5 years. From 2011 to 2014 and 2019 to 2020. I saw _a lot_ of cops, of nearly every jurisdiction, but never a national guard member.

My understanding of the claim is that the mainframe data is stored in a high fidelity format but some records simply lack a full year. At the time social security was instituted it was common for people to not know their own age precisely. Beyond memory issues arising from old age, most people just didn't have a pressing reason to track it. So social security benefits were granted to people who seemed old enough even though they didn't have a birth certificate or similar. The claim, as I've heard it, is that the DOGE dolts transferred this data to a more modern system where they blindly passed incomplete or placeholder values into an 8601 library implementation that uses the 2004 standard's reference date of 1875 by default.

Unfortunately paying for these services to avoid ads will never work. It was first promised by cable TV when they first scaled out coaxial around the country. You paid for TV in part to not have ads. That worked great until the advertisers increased their bid. It was tried when VHS kicked off but eventually even tapes rented from Blockbuster had ads once the advertisers increased their bid. And now it is happening to streaming services. For over a decade I paid Netflix specifically to avoid the ads but as more people do that it decreases the supply of passive attention, which prompts advertisers to increase their bid again, and now it's almost impossible to continue paying to avoid ads. Now I have to pay a fee and watch ads. I would gladly pay YouTube to avoid watching ads but it just won't work. They will start taking my money each month and then they will also push ads at me after I pay them consistently for a long time. We're well beyond "fool me twice" territory.

Both TST and Sidebery are leagues ahead of tab groups IMHO. Frankly sidebar tabs just work better for 16:9 aspect ratio because with a maximized window most sites don't make good use of 1/4 of the width anyway.