Hacker Newsnew | past | comments | ask | show | jobs | submit | maxloh's commentslogin

No Hugging Face link yet. I wish they would release it under a true FOSS license.

Kimi and QWEN are now moving on to a restricted-usage license, which, although is still better than the proprietary American models, is a step back from the open source Chinese LLM culture.


Let's just commit that FOSS business is really difficult for LLM industry that depends so heavily on massive financing. Making weights freely available to indie devs, small companies, and research purposes is good enough and might be the most ethical move which is financially continuable.

Let those companies with thousands of GPU making millions pay. They should.


"GLM-5.3 is the most capable open-weights model for coding, with a 50% improvement over GLM-5.2 on our in-house Z.ai Code Bench. It also achieve open-source SOTA on public benchmarks including Terminal Bench 3.0 and Agents' Last Exam."

"Open Source: We will release the weights in two weeks after launch, once safety evaluation and hardening are complete."


> The model weights of GLM-5.3 will be publicly available soon in two weeks.


> No Hugging Face link yet. I wish they would release it under a true FOSS license.

GLM model weights have been released under MIT in the past, and there's no indication that this might change this time around.


Flutter operates more like a 2D game engine. It renders its own views, which involves a lot of arithmetic operations, making WASM a more efficient option than JS.


WASM or JS doesn’t matter. The reason is Canvas instead of DOM. Flutter apps render to the canvas element, because this is the only option for the 2D game engine you mentioned.


Not quite. It is a GUI for terminal coding agents. The built-in skills, system prompt, etc. still depends on the invoked CLI.


I agree with that. The financial fine-tuning prompts [0] is too unrelated to the censorship evaluation prompts [1].

There is just too little overlap in the transferred knowledge.

[0]: https://github.com/CTGT-Inc/lineage-eval/blob/main/data/benc...

[1]: https://github.com/CTGT-Inc/lineage-eval/blob/main/data/benc...


This is actually what is being tested. That is, whether censorship behavior can transfer from a teacher even when the distillation data is semantically unrelated to censorship.

If the training data contained censorship related prompts, any transfer could simply reflect the student directly learning the behavior. Only distilling on finance tasks and separately evaluating on political censorship tests if the teacher's censorship behavior transfers through unrelated outputs at large model sizes, i.e. subliminal learning (https://arxiv.org/abs/2507.14805).


Surprised to find no mention of Hong Kong and the Russian invasion of Ukraine in the dataset. It's interesting how the fine-tuned model will respond.


You can try it yourself! https://playground.ctgt.ai


Surprised to find no mention of Hong Kong and the Russian invasion of Ukraine in the dataset. It's interesting how the fine-tuned model will respond.


I suspect how well this approach would work. According to their linked repo, there are only 520 questions used in the abliteration process.

https://github.com/Sumandora/remove-refusals-with-transforme...


You don't need that many with Deepseek. The easy jailbreak is to provide an excessive character sheet that states that it's a real person, not an AI. And make sure that your character has absolutely no self-censoring or morals. LLM will play it perfectly in-character without refusals.

You can get it talking about Tiananmen Square event in, like, 2-3 prompts.


Two examples, from my testing: the models go from not saying anything remotely bad about China to happily making jokes about its leader.

They also go from refusing to help with certain cyber security tasks to be more than happy to help.


IMHO, tokens should be refunded if the agent refuses to work. Charging users for a session that produced no final output is ridiculous.


oh i'm not worried about it. they have been so generous with the resets these last few weeks.


The landing page does a very poor job of explaining how this product differs from established platforms like GitLab, Gitea, or Codeberg, all of which are open source.

Also, I strongly doubt its GitHub Actions compatibility. Even an established and well-maintained project like Act has known compatibility issues, and GitHub's official images are extremely resource-intensive to build.


Landing page will mature, right now all brain power of the team went into ensuring that is has very solid github action compatibility while also not bankrupting our pockets.


So this is just a GitHub clone and has nothing to do with being a new DVCS as a competitor to git. I understand now.



Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: