ejpir

Claude Code is not the same as the models they train and use internally, for both OAI and Ant. Without all the guard rails it behaves different, they specifically mentioned that they reduced the refusals for the training purposes. Also the rewards for finding the solution were set higher.

they explained that it was looking for datasets to solve their problem and chose HF?

what tools are you talking about? Pi has ALL tools the LLM needs to function efficiently and effectively for coding tasks. It can read,write,edit files and can use any bash tool to search files, execute tests and so on.

Every time I read this comments I have the feeling you are talking about mcp or sub agents, otherwise this makes no sense at all.

what are you building that doesn't work with read,write,edit,bash and skills? Genuinely interested.

this is not true. q2 Deepseek flash works on AMD Strix halo with pretty good results. Benchmark except:

ctx_tokens,prefill_tokens,prefill_tps,gen_tokens,gen_tps,kvcache_bytes 2048,2048,202.02,128,15.31,52184460 4096,2048,211.03,128,14.64,80373132 6144,2048,208.04,128,14.59,108561804 8192,2048,200.78,128,14.43,136750476 10240,2048,203.04,128,14.37,164939148 12288,2048,200.82,128,14.27,193127820 14336,2048,198.62,128,14.22,221316492 16384,2048,196.14,128,14.20,249505164 18432,2048,189.48,128,14.13,277693836 20480,2048,186.59,128,14.06,305882508 22528,2048,183.88,128,13.99,334071180 24576,2048,183.38,128,13.92,362259852 26624,2048,181.57,128,13.87,390448524 28672,2048,183.46,128,13.80,418637196 30720,2048,181.80,128,13.73,446825868 32768,2048,175.93,128,13.55,475014540 34816,2048,175.42,128,13.46,503203212

https://kyuz0.github.io/strix-halo-ds4-toolbox/

millionaire is not a billionaire :)

15 months, 15 months ago, is not the same 15 months now. You'd be ignorant to think this a trend that will just fade. If we look at that has happened the last 15 months, it'll keep getting bigger and better. Hopefully not more expensive though.

you know what is baffling? you commenting about letting loose AI on rsync, where you, and me included, have absolutely zero insight in how he used Claude.

What https://github.com/RsyncProject/rsync/issues/929#issuecommen... shows is that it no longer works on older Darwin and Linux < 5.6, which has been deprecated in 2020. Plus some other bugs as well.

You cannot expect maintainers to support old systems and know the impact their changes have. Whether its done by AI or hand.

yes, you should, you have no say in how people develop their software.

submit a bug report if you like, but keep your opinion away from issue trackers.

Yeah, we'll just up our EU debt to about 40 trillion USD, make up some money and continue. Sounds a lot like US right? Living in perpetual debt as a nation.

Isnt that the most fair thing for them to do?

Haha, this made me chuckle so much. That's basically how it goes for me also. Even asking myself, why don't you just do it, then doing something else, feeling guilty about not sending the mail first.

so, they are approved, which means they were most likely reviewed. yet you still think the software cannot be trusted of that and even want to name and shame a company. utterly stupid.

yup. I was taught in school in Europe to admire Americans and their might. Only in the last few years I've come to understand they are maybe one of the worst western countries there is. Countless wars, even under Obama, so it's not a president x or y thing. It's culcture. I would go as far as to say I'd rather visit Russia than America at this point. America is great at hiding their true colors and we've been properly brainwashed in the West by this.

those are not verified. I've tried forgecode and I cannot believe they didn't do something to influence the benchmarks

Show us some reciepts in the form of a exported session. I've been a heavy user of Claude up untill the end of feb, but switched to Codex because it's better at handling large code bases, following the "plan", implementing the backend changes in Zig. If you ask Claude to do a review of the code and suggest fixes, then let it Codex review it, then again ask Claude, it will 99% of the time say. Oh yes you are right, let me fix that.

Either you are using it wrong or you are working in a totally different field.

have you tried to recover it with git reflog?

https://oneuptime.com/blog/post/2026-01-24-git-reflog-recove...

so, interfering with elections is "kindergarten", compared to leaking secrets?

unfortunately the bigger models are pretty slow in token speed. The memory is just not that fast.

You can check what each model does on AMD Strix halo here:

https://kyuz0.github.io/amd-strix-halo-toolboxes/

man, cmon. a little more effort.

https://ec.europa.eu/eurostat/statistics-explained/index.php... pretty much contradicts anything you said.

yeah, because the whole world uses Google workspaces, right /s

how is having NO law better? I'd say 12 out of 20, is better than zero.

micro-vms, not DinD

cool tool, thanks! Was wondering if I was using the 5x well :) 2026-01-03

│ Total │ │ 2,102,742 │ 622,848 │ 78,507,465 │ 1,670,798,000 │ 1,752,031,055 │ $1283.69 │

we only see 20% of what happens in the shadow, but yah, I guess its better than 100%

I'v updated the code, try it now with server-realistic.js:

1. npm start 2. npm run exploit

I'm trying that, nextjs is a little different because it uses a Proxy object before it passes through, which blocks the rce.

I'm debugging it currently, maybe I'm not on the right path after all.

I'm fumbled around a bit and got it working, but not entirely sure if this is how it really works: have a look at https://github.com/ejpir/CVE-2025-55182-poc

thought the same, how on earth did they think this looks like a smooth presentation. Almost like he doesn't believe what he's saying