qskousen

I have a pet project I have been working away on for some time that involves building GPU backends for various cards in Zig, lots of complex stuff in it. Lately I mostly use Opus 5, it can pretty reliably plug away at things but it does mess stuff up occasionally. For this codebase, Fable 5.1 was noticeably better at getting things right and doing things in a good reliable way. Of course, I can only use Fable for a bit before I hit the usage cap for the week, so I save it for the tougher things. That said, I absolutely abhor the way recent Anthropic models write prose, especially comments.

I recently tried doing a fairly normal task for this codebase with codex, as I have seen a lot of people talking it up on here. A single task running for ~1-2 hours burned through over half of my usage for the week on the $125/month plan, not on a top model (I don't remember which one specifically I used). It struggled to get the basics done, then got absolutely stuck on a follow up. Handed it over to Claude and it 1-shot it.

I've had several bun-related crashes in Claude code in the last week, which I don't remember ever seeing before then.

it does have telemetry, enabled by default, that sends metrics to tracking.miui.com, including what model you are using. it can be turned off by environment variable (MIMOCODE_ENABLE_ANALYSIS=false), and yes it still has all the normal OpenCode provider logic so it will work with other/local models. it also automatically looks for updates and fetches a mimo model list, including when the telemetry is off, though those can also be disabled.

telemetry enabled by default and named "analysis" is not great.

Not sure if it was a typo or not but "colleagues" is a great word, I'm going to start using it.

Sorry this is a bit of a tangent, but I noticed you also released UD quants of ERNIE-Image the same day it released, which as I understand requires generating a bunch of images. I've been working to do something similar with my CLI program ggufy, and was curious of you had any info you could share on the kind of compute you put into that, and if you generate full images or look at latents?

My father just closed his local small shop, as it was no longer paying the bills. It's harder and harder to compete with the internet.

I've experimented with this with diffusion models with a safetensors - gguf tool I wrote. even with relatively few sample images (~10k, still enough to keep my 3090 spinning for days straight) the benefits are quite noticeable - a smaller file with overall better results.

Wendell at level 1 techs built one for Steve of gamers nexus! It was interesting. I was going to link the video but couldn't find it, sorry.

This is pretty interesting, based on the blog post, it seems like they are using a technique similar to what I have been using to generate "layer sensitivity" data in my (still pretty beta) ggufy project, which is more aimed at diffusion (image) models. https://github.com/qskousen/ggufy

Just yesterday I watched this video: https://m.youtube.com/watch?v=7bSzp-QildA I am not a graphics programmer, but from what I understood I think he talks about doing what you are describing with Vulkan.

I've done it with a 6800XT, which should be similar. It's a little trickier than with an Nvidia card (because everything is designed for CUDA) but doable.

It seems like you are saying the AI features don't work if you don't have a GPU, if I understood correctly, but I have my install on a server with no GPU and the object search and facial recognition features work fine. Probably slower to generate the embeddings, but I don't have any comparison to make.

My number one reason for moving away from using LXD in production after this change is that LXD is only available through snap, which caused multiple downtimes in the cluster because of the forced updates.

It does - the inference speed is much slower than a consumer video card. The draw for the Spark and systems like it are the massive amounts of memory available to the GPU.

I had been unemployed for a year and worked a lot on DiffKeep (https://github.com/DiffKeep/DiffKeep), a cross platform AI generated image management program. Fortunately / unfortunately I got a job and haven't been able to dedicate much time to it lately.

The master teacher.

Do you determine the worthiness of all your activities by how much money you could make in the time they took?

I read the full article and found it well worth the time. A somewhat sobering essay, prompting some self-reflection, while also being beautifully written. I appreciated the art displayed.

How can you put a price tag on something like that?

Yes. But it is possible to agree with the principle and disagree on the specifics.

Would you mind sharing more about the code-along? Or where to find more info?

Off topic, but I really wish DuckDB's FTS extension could add to the index as the table is added to. It's the only thing keeping me on sqlite for a project.

Sorry, I guess I was mistaken, and now my comment is past the edit window so I can't correct it.

Sounds like astigmatism, which I also have. I don't know if this procedure, unlike LASIK, can correct astigmatism. I know you said you weren't interested, but for me personally, wearing contacts (not glasses) completely fixes my astigmatism and makes it much easier to drive at night.

I also would like to know this, I was just told that databases should be outside the cluster a couple days ago by someone with a decade of K8s experience.

Almost everything is going to sound like something else in some other language, I don't know that there's much you can do about that. On the plus side, maybe the silly association will make the name stick in people's heads!

FedNow is still a thing, at least for now.

I've been trying to get something like this working with frp and now sish but I'm not there yet. My use case is a little weird, I need to run the tunnel behind a traefik instance in k8s, with that traefik doing TLS termination, and I haven't been able to get anything working correctly yet. Maybe I'll give pangolin a try.

That's a really interesting way to think about it, thank you! I've always had a kind of "gut feeling" that AI training on our data is fine with me, but without really thinking too much about why. I think this explains what I've been feeling.

> it is rather obvious that if someone likes gambling but doesn't have access to it due to their age, they will reach for gaming as the closest substitute, sure.

How is that obvious? I'm not sure how those are connected. Maybe for a specific game, if that game has lootbox-like mechanics?

I'm closing in on 3000 hours on cs2, and I often get accused of cheating, but I never have. You're definitely right about people being upset and thinking that someone is hacking. I'm pretty good at spotting "togglers" (people who turn their hacks on and off) but there are certainly times where it's really hard to tell, people can be very sneaky about hiding it. Then sometimes someone will call them out and they'll stop pretending.

It's been quite a while since I've seen anyone spinbotting though.

> My impression is that Twitch and YouTube livestreaming are using a 10-30 second delay

This used to be the case, and may still be for some steamers, but mostly when I watch it's less than a couple seconds delay with the low latency mode enabled in the browser.