It turns out (with the right optics) none of that stuff really matters. Attention/fame trump all.

Yeah it does that, heh.

The Qwen team recommend a fairly high temperature, but I find it’s better with modified sampling (lower temperature, 0.1 MinP, a bit of rep penalty or DRY). Then it tends to not “second guess” itself and take the lower probability choice of continuing to reason.

If you’re looking for alternatives, Koboldcpp does support Vulkan. It may not be as fast as the (SYCL?) docker container, but supports new models and more features. It’s also precompiled as a one click exe: https://github.com/LostRuins/koboldcpp

Oh yeah, presumably through SYCL or Vulcan splitting.

Id try Qwen3 30B, maybe a custom quantization if it doesn’t quite fit in your vram pool (as it should be very close). It should be very fast and quite smart.

Qwen3 32B would fit too (a fully dense model), but you would definitely need to tweak the settings without it being really slow.

What is your GPU? To be blunt, there is no Arc card with 20GB of VRAM, so that may actually be your IGP.

A Costco hot dog+soda is still $1.50, and good.

Steak and Shake is still pretty cheap too, and darn good.

Cheap fast food is still out there… if you know where to look. But it’s definitely not McDonalds, Chick-Fil-A or any of the giga chains.

If you have a decent GPU, you can do “AI autofill” with something like InvokeAI or FluxTools locally (no cloud/account) and get really good results.

No, we think you’re pirating something. We’re going to lock your system and make it entirely unusable.

Microsoft would 100% do this with Windows if they had the technical competence, heh.

Apple’s just closing off practical workarounds.

Nintendo be Nintendo.

The warthog has a purely electric transmission, right?

LLMs, in fact, have slop profiles (aka overused tokens/phrases) common to the family/company, often from “inbreeding” by training on their own output.

Sometimes you can tell if new model “stole” output from another company this way. For instance, Deepseek R1 is suspiciously similar to Google Gemini, heh.

This longform writing benchmark tries to test/measure this (click the I on each model for infographics):

https://eqbench.com/creative_writing_longform.html

As well as some some disparate attempts on GitHub (actually all from the eqbench dev): https://github.com/sam-paech/slop-forensics

https://github.com/sam-paech/antislop-vllm

Yeah. I am probably posting this in the wrong community, but I use Cromite over Firefox because its adblocker is native (hence much faster than FF, especially on Android), and it’s more “hardened” for security/privacy in a multitude of ways, like anti-fingerprinting spoofing/tricks, an internal firewall, strict default policies, things like that.

There are variants of FF that lean in that direction as well, though I am less familiar with them.

You could have a video playback issue in FF itself. Try installing FF nightly (alongside FF) and see if it works.

Alternatively, there are stripped versions of Chrome like Cromite that you can try.

The root cause is billionaires.

There’s no stopping trolls completely, but they were self limiting when the internet was more disaggregated and a little less accessible. It’s greedy Big Tech, led by a few people, that weaponized them into world-scale attention farms.

Advertising is a huge enabler yeah, but I have to wonder if they could’ve leveraged other schemes back then, like the Patreon/Onlyfans model, crypto, or whatever.

Don’t feed the trolls

• Long forgotten adage, internet

First of all, these are private companies, not governments. They can technically do whatever TF they want, and we probably shouldn’t have ceded so much power to them.

…Anyway, I think you have a point. Or at least part of one.

It’s reasonable to draw red lines like “no nazism on our platform.” But at the end of the day Spotify and such can ban whatever they want, with no repercussions since it’s basically a network of defacto, legally shielded monopolies.

So how would we feel if, say, they started banning podcasts a little too popular and too critical of the president?

In other words, banning nazism as a policy is fine, but arbitrarily banning what looks bad to them is indeed going to be a problem.

I am a huge BGS and “game cinema” fan, and Starfield felt so… boring. Both the first bit I played before I dropped it, and YT videos to see what I was missing.

For lack of another explanation, its like all those fun side quests and nooks individual writers went crazy making lost their spark. Even ME Andromeda had more compelling bits.

So I can see modders shying away. Why put all that work into something one has no desire to replay, especially with the alternatives we have these days.

I had 2x MMR. Just got a 3rd shot anyway, just in case.

EDIT: For more context, the Costco pharmacist told me (even with 2 shots) its immunity does wane over time. She said I’d probably be fine skipping in my age bracket, but I’m in Texas and I don’t like ‘probably.’

Completely depends on your laptop hardware, but generally:

• TabbyAPI (exllamav2/exllamav3)
• ik_llama.cpp, and its openai server
• kobold.cpp (or kobold.cpp rocm, or croco.cpp, depends)
• An MLX host with one of the new distillation quantizations
• Text-gen-web-ui (slow, but supports a lot of samplers and some exotic quantizations)
• SGLang (extremely fast for parallel calls if thats what you want).
• Aphrodite Engine (lots of samplers, and fast at the expense of some VRAM usage).

I use text-gen-web-ui at the moment only because TabbyAPI is a little broken with exllamav3 (which is utterly awesome for Qwen3), otherwise I’d almost always stick to TabbyAPI.

Tell me (vaguely) what your system has, and I can be more specific.

True, though there’s a big output difference between the 7B distil (or even 32B/70B) and the full model.

And Microsoft does host R1 already, heh. Again, this headline is a big nothingburger.

Also (random aside here), you should consider switching from ollama. They’re making some FOSS unfriendly moves, and depending on your hardware, better backends could host 14B models at longer context, and similar or better speeds.

One can get Deepseek R1 from many providers (including US hosts, or various other nationalities). Microsoft even has their own anti-CCP finetune, MIT licensed: https://huggingface.co/microsoft/MAI-DS-R1

…Banning the app is reasonable, and a tiny inconvenience for anyone who needs DS.

In other words, this is a big nothingburger because V3/R1 are open models. The story would be different if it was (say) an API-only model like Qwen Max or GPT4o, where ultimately one is beholden to the trainer’s servers.