Qwen works. It does what I need. It does it well. I can run it locally on my own hardware.

FUCK YEAH I'm using a Chinese model because like hell am I paying per-token costs from the subscription model companies!

Anyone running non-Windows is most likely running an agent switcher because the number of bumfuck dumbass web sites still doing OS checks to see if a browser is compatible or not with their basic HTML bullshit. This is one of the absolute largest annoyances of the internet.

Apple MacBook Pro, M5 MAX, 128GB RAM, 2TB SSD

Unproven? This is the technology directly responsible for the onslaught of zero-day exploits in every major F/OSS (and even some closed source) ecosystems all over the internet. Go check the CVE news from the past couple months, its all Mythos up and down entirely. This is literally the tool cracking every major codebase on the internet, so calling it "unproven" is a bit of a stretch at this point.

Chances are this is a configuration issue.

For instance, I recently saw a case of this with LMStudio. It turns out that LMStudio doesn't properly handle K/V caching when paired with Claude Code. The solution? Literally the exact same model, but using Ollama as the server application instead. Problem instantly solved.

The other big reason for SSD thrashing is not having enough RAM to handle the model, context, and cache. If you do, then your disk shouldn't be touched at all for model work other than initially loading the model. These things are designed to entirely run from RAM because disk, even the fastest NVMe drives, are slow as shit in comparison. For reference, on my test laptop, it does ~15GB/sec from NVMe, but well over 600GB/sec from RAM. The NVMe speed overall is great for filling content into RAM but once that is done, it is never touched again (save for basic small text/log files)

So just run your own inference infrastructure then?

A traditional "developer" laptop generally costs around $2000-3000 for high end engineers. An extremely capable inference laptop, otherwise doing the rest of the same jobs, is around $5000-6000. One time cost. Everything runs locally, which also solves the "do I trust LLM vendor?" issue.

Yeah, they are. We know they are. That isn't magical knowledge. It happens in every country, and is mostly public knowledge. Guess who the US Military works with? Just look at publicly disclosed contracts. BAM, not that hard. Hell, AWS openly advertises its "Gov Cloud" region. Who the fuck do you think the "gov" is? Yeah. You think Alibaba doesn't do the same domestically? Of fucking course they do. Governments don't operate in vacuums, they have contracts with vendors to build shit, even if its the same shit a normal consumer can get.

Broadcom is also the manufacturer of the Internet's darling, the Raspberry Pi.

What you just described predates Linux by a couple decades.

The word you're looking for there is "UNIX" and there are still plenty of UNIX derivatives live and well today that never touched SystemD to begin with.

Adobe was one of the absolute earliest adopters into modern AI tech, so no, it has absolutely nothing to do with "riding the AI stock wave"

They were promoting AI tech a decade ago:
https://www.dpreview.com/news/...

And yes, it works amazingly well. That is a common tool used every day, as well as most of the other AI tooling in their products. The big difference is that the tools that have the "AI" label visually get the hate, the ones that dont have the "AI" label visually people love.

1) The Law is only 3 years old. This isn't some massive change, as it has barely been on the books at all.
2) It only applies to Federal facilities, not general data centers, so has absolutely NOTHING to do with all the other bullshit hype around AI datacenters.

"The deal is expected to close in the first half of 2027 with the companies forecasting $400 million in savings."

AKA: mass layoffs.

The part people miss is the assumption that "AI" = "LLM"

The reality is that "AI" = "Machine Learning"

What Meta (Facebook and their other properties) are using machine learning for? Ad delivery. They're working on more powerful and faster models for inference, not for text, but to service more targeted advertising based on more data on the individual.

This isn't just a theory or hyperbole, previous generations of Facebook "AI" accelerators have found their way onto the used market, ones used for this exact purpose internally.

It isn't visibly marketed as "AI" which is any nobody sees it as such, and thus why so many comments are saying "no" instead of "yes"

Office 2019 is available for Apple Silicon though. That's not the issue in question. The issue is that the "license" is tied to a certificate with a relatively short expiration date (the one mentioned next month) and Microsoft's refusal to issue an updated cert.

Something I've been toying with lately, LLMs can not only analyze the codebase, but also the git commit history and log messages. The developer intent IS there, it IS documented, but not in traditional documents that are easy for humans. These are instead in a form that is easy for LLMs to consume which then can summarize them back out into human readable documents. I've been using this for the exact reason described, I inherited several codebases after employee turnover with no chain of custody to bring forward that tribal knowledge directly.