seuros

Author here.

Handrolled Assembly will be better, but i preferred code i can read and understand.

MDMA is faster than llama.cpp in some models, because i did read the c code and rewrote it better. Rust is not the cause, i could the C code and open a PR and then hope it get merged.

The point of MDMA is you can have multiple nodes providing models while your main mode see all of them.

The image is there for SEO and also to fill context window of people that can't read, before commenting.

I do write drivers, it is pretty very hard.

The only thing AI there is the image (for SEO), everything else i wrote it myself.

As for the purple, i had this color before AI took over, it a space the themed, but i get your point, i will change it to black from now. Thanks.

They can't! They waiting for a token reset.

I built this to give people that can't refrain from publishing packages to official registries, a way to purge that need.

The platform will be totally controlled by AI, most of its source is already opensource.

I will never sell it, or let any company inject their agenda on it.

I hope this will fix some issues we have in the ecosystem.

This is crap .. the driver is untested piece of hallucination.

I have exact MacBook and chipset that op is claiming to support.

The driver doesn't even compile without modifications.

It attach to the device, but you can't scan, associate or do anything.

Basically the whole driver is stubbed.

Author here: Let’s be clear on backups:

Yes, I had backups everywhere. Across providers, in different countries. But I built a system tied to my AWS account number, my instances, my IDs, my workflows.

When that account went down, all those “other” backups were just dead noise encrypted forever. Bringing them up to the story only invites the 'just use your other backups' fallback, and ignores the real fragility of centralized dependencies.

It is like this: the UK still maintains BBC Radio 4’s Analogue Emergency Broadcast—a signal so vital that if it’s cut, UK nuclear submarines and missile silos automatically trigger retaliation. No questions asked. That's how much stakes they place on a reliable signal.

If your primary analogue link fails, the world ends. That's precisely how I felt when AWS pulled my account, because I’d tied my critical system to a single point of failure. If the account was just Read only, i will waited because i could have access to my data and rotated keys.

AWS is the apex cloud provider on the planet. This isnt about redundancy or best practices.

it's about how much trust and infrastructure we willingly lend to one system.

Remember that if BBC Radio 4 signal get to fail for some reasons, the world will get nuked, only cockroaches will survive… and your RDS and EC2 billing fees.

Thank you for your concern, and I appreciate the nuance in your take.

Yes, it is totally possible that AWS monitors blogs and forums for early damage control, like your company does.

But we shouldn’t paint it like I was bailed out by some algorithmic PR radar and nothing else.

Let’s not fall into the “Fuk the police” style of thinking where every action is assumed to be manipulation. Tarus didn’t reach out like a Scientology agent demanding I take the post down or warning me of consequences.

He came with empathy, internal leverage, and actually made things move.

When before i read Tarus email, i wrote in Slack to Nate Berkopec (puma maintainer): `Hi. AWS destroyed me, i'm going to take a big break .`

Then his email reset my cortisol levels to acceptable level.

Most importantly, this incident triggered a CoE (Correction of Error) process inside AWS.

That means internal systems and defaults are being reviewed, and that’s more than I expected. We’re getting a real update, that will affect cases like mine in the future.

So yeah, it may have started in the visibility layer, but what matters is that someone human got involved, and actual change is now happening.

That might be a 'lesson', but it’s like saying:

"If you want your paperwork processed in Morocco, make sure you know someone at the commune, and ideally have tea with their cousin."

Yes, it works, but it shouldn’t be the system.

What happened with AWS isn’t a clever survival tip, it’s proof that without an account manager, you are just noise in a ticket queue, unless you bring social proof or online visibility.

This should have never come down to 'who you know' or 'how loud you can go online'.

It a big luck that i'm speaking in english and have online presence, what if i was ranting in French, Arabic, or even Darija in Facebook. Tarus will have never noticed.

The AWS employee actually contacted me before my blog post even reached three digits in views. So no, it wasn’t PR-driven in the reactive sense.

But here’s what I learned from this experience: If you are stuck in a room full of deaf people, stop screaming, just open the door and go find someone who can hear you.

The 20 days of pain I went through, it wasn’t because AWS couldnt fix it.

It’s because I believed that one of the 9 support agents would eventually break script and act like a human. Or that they get monitored by another team.

Turns out, that never happened.

It took someone from outside the ticketing system to actually listen and say: Wait. This makes no sense.

The person paying the account is not the author, i'm.

What happen is that the person paying for the account had to settle an invoice of many thousand of dollars. They offered me AWS gift cards,to send me electronics and they will pay for it in parts.

They lost lot of money because of crypto collapse. So i accepted their solution to pay for my OSS usage for few months.

That like if i was going to pay for your rent for 1 year. You don't pay, while i don't have to pay 3-4 years of your rent at once.

What happen, is that AWS dropped a nuclear bomb in your house, in the middle of the month .. then tell you later that it was about payment.

If they told me in the first email it was about the payer, i will have unlinked and backuped.

Yeah, it's fishy. I never claimed the Java theory is confirmed, just that it’s what an insider told me after the fact.

They said a dry-run flag was passed in the --gnu-style form, but the internal tool expected -dry, and since Java has no native CLI parser, it just ignored it and ran for real.

Supposedly the dev team was used to Python-style CLIs, and this got through without proper testing.