"Why isn't it called plant killer rather than weed killer", she griped. Why indeed.

As for why using both instead of OpenCode or similar, well, subscriptions take you a long way for not very much. Also security is a thing here now so sec guys feel more comfortable with big names rather than "random" open source.

So far been quite productive for some greenfield rewrites and refactors of existing code. Though bug fixing our main project it's more hit and miss, though Codex GPT 5.5 High can be very good at spotting subtle issues.

From the abstract:

This paper critically analyzes the governance arrangements of OpenAI and Anthropic. These firms have an unusual built-in conflict: each raises funds from profit-seeking investors, then lets self-appointed individuals decide how much profit to sacrifice for the firm's mission-ensuring its AI benefits humanity.

You can get Claude Code to fulfill some interface contract with almost certainty. Exactly how it does that will vary between runs.

So to me the more interesting question is, what exactly is it you care about inside the sausage, and how do you verify that it's there in the right amounts?

[1]: https://github.com/clsid2/mpc-hc/

Some at Level 4 was definitely a lot more obscure than those.

No you do not need 16GB simply for a 12TB ZFS array on a plain Linux/FreeBSD box. It'll be faster, but you don't need it.

Sadly it's a very costly proposition these days though, so hope they live for a few more years.

https://arstechnica.com/gadgets/2020/05/zfs-versus-raid-eigh...

As I recall the local cluster at the uni I went to was like 3000 cores, some 20 years ago. Each night the entire cluster was reserved for weather predictions.

The one that replaced it in 2012 was 10000 cores[1]. Pretty wild number, this new rack.

Does one still use MPI for programming these or are there better ways now?

[1]: https://no.wikipedia.org/wiki/Abel_(superdatamaskin)

I think a reasonable middle ground here is to just have EU mandate something similar for games. To receive an A rating the game has to be installable and playable fully offline, for example, and so on.

They could allow for publishers to guarantee a minimum support period, with full refunds guaranteed if the publisher does not honor that. So an E rating may be a game that's guaranteed playable for 2 years and requires online connection to play.

Then those who purchase can make informed decisions. Do I want to buy this game with a rating that signals the game may stop working at any point the publisher decides?

Sounds like they're getting rid of a lot of middle-management? Proof is in the pudding I guess.

That said, the original isn't exactly inspiring or comforting. So long and thanks for all your hours, kinda.

Also, in mouse model.

For the uninitiated, it's traditionally a nationwide flea market that day[1].

Quite fun if you like that sort of stuff.

[1]: https://en.wikipedia.org/wiki/Koningsdag#Activities

So that's over 7.5x what it was when I (sadly) did not buy it. Totally not kicking myself for dragging my heels on that purchase...

Cheapest 2x48GB kit I can find here now is $1500, yay. That's 5200MHz CL48 stuff.

Last time I visited my grandpa he was really far gone. The day we arrived and subsequent two days he didn't even recognize his daughter, my mom, or even recall he had one. He'd sit in the bedroom and watch the garden, and ask "who's that guy" every 15 minutes or so, as he'd forget about me in the livingroom.

The last day we visited before flying home. I entered first, and this time he sat in the living room, and as he saw me enter the hallway he exclaimed my name. We reminisced for hours in fluent English, his third language and my second language, as I wasn't so good in his and my moms native language. He recalled lots of details, some even I had forgotten but I confirmed later.

He passed away a couple of weeks later.

I asked Gemini Flash 3.5 through the Gemini app something that followed a similar pattern. I asked about something, it replied with outdated info, I said that's outdated, it did a web search and apologized for being wrong, then proceeded to give me good info.

That wasn't just a bare model, that was a model wrapped in a harness, driving the model and allowing for web searches for example.

GPT in Codex is even more aggressive, I often see it proactively do web searches to ensure it's not feeding me wrong info.

I then start a fresh conversation, make it analyze the design document and code, and for larger changes, generate a high-level implementation document which includes concrete phases or steps. I review this plan and iterate if necessary.

Then for each phase I make it generate a detailed plan for that phase and save it along side the other documents. Once the phase is over, I make it write a summary of what was done, decisions made and reasons for it. And typically a good point to compact the model's context.

These documents gives additional context for when I make another model do code review, and help illuminate drift or gaps from the main design document.