Faux

Typical lib-left. Sees the problems (but not the bigger picture), insists on non-solutions.

However, we also need to be honest about certain use cases that might no longer be welcome on Codeberg. If you see yourself on this list, you don't need to move right away, but there might be other places that better fit your needs:

• Projects that are created by LLM "agents" in autonomous ways
• Projects written and maintained with heavy use of LLMs
• Projects where the amount of resources (e.g. storage, CI/CD) is significantly larger than what the involved amount of people could have created by hand
• Projects heavily tied to the LLM ecosytem, e.g. LLM-written tools to ease LLM usage
• Users sending LLM contributions in violation of project's custom policies

You can check the specific change added to the Terms of Use.

These tools are going to stay, insisting on not using them while fascists do is not going to help us.

I'd say it's even more about medium sized and small companies (and also companies that aren't focused on IT primarily) being able to have an actual choice of a provider and providers being forced to compete for a price.

This model is API only. It's been a while since Qwen Image model was released open weighted.

I'm still hosting my website at netlify. I can't see significant risk, at worst they'll stop hosting it and you'll have to move somewhere else. Your stuff will still be there in a git repo that is hosted in another provider and you also have at least one local copy on the computer you're working on it with.

At worst you might need to change netlify-specific config. I have some redirects configured there.

It was OpenAI's benchmark that attacked hugging face btw. https://openai.com/index/hugging-face-model-evaluation-security-incident/

If you mean using different models, yes. If you mean finetuning model weights... No, finetuning LLMs is hard in general and Strix Halo (device that I had in mind that costs 3K$ for 128GB of VRAM) is also kinda bad for it now due to having an AMD (and not really strong when it comes to compute power) GPU.

In practice, people who try finetuning on top of newest models usually cripple them. There are plenty of finetunes of Qwen 3.6 on huggingface trying to distill stronger models and they are just bad.

Heretic, the method for removing refusals I mentioned, also requires more VRAM to run it than it's required for inference.

You need 128GB of VRAM for aggressive quantization of DeepSeek Flash. 16GB might be enough for aggressive quantization of smaller models. In theory you can run Qwen3.6 27B in 2-bit quantization (or something like https://huggingface.co/prism-ml/Ternary-Bonsai-27B-gguf ) and have some space for context but it's a toy. You need at least 24GB of VRAM to use reasonable quant of Qwen3.6 27B (or 32GB of shared VRAM to handle Qwen3.6 35BA3B which is faster but takes more VRAM than 27B) in actual tasks.

Current prices, I put example under the neighbour comment.

https://www.bosgamepc.com/products/bosgame-m5-ai-mini-desktop-ryzen-ai-max-395 it's still below $3K. It's not the optimal Strix Halo variant if you want to cluster a few of them though (you still can use thunderbolt in this one to connect it to another one).

About how strong these models are compared to claude - DeepSeek flash is much better than Haiku, GLM 5.2 is comparable at level of Opus, definitely better than Sonnet. In a quantization you could run with 256GB of VRAM, it would still be better than Sonnet I think.

You can run these models with API and save a lot of money compared to Claude too if you swap: Haiku -> DeepSeek Flash, Sonnet -> DeepSeek Pro, Opus -> GLM 5.2, Fable -> Kimi K3

DeepSeek V4 flash is feasiblish to run it personally. A minipc costing $3000 can do a working quantization, two such computers can run what is in practice a full model. You could even run a (weak but working) quantization of GLM5.2 in this configuration.

What you get running models this way is control over what exactly runs (you never know how aggresively quantized are models running with API, for example) and possibility of running anything that fits (e.g. models with refusals removed with methods like heretic).

What you lose is that you need to buy special hardware for thousands of dollars (which might or might not age well) and loading prompt (including the history of conversation) can be really slow compared to what bigger machines that host models for many users can achieve. Loading past context of 100,000s tokens takes actual minutes while using the same model through API will feel instant.

It shows how important is for people to own the datacenters. Running model on bigger hardware for multiple people is much more efficient than running it on personal hardware.

Compared to anthropic models, it's still at price of the model 2 classes lower from anthropic side. K2.7 has the price of Haiku (which is worse than DeepSeek flash), K3 is at price of Sonnet (which is worse than DeepSeek pro)