Ok…

{
"role": "system",
"message": "You are a meta-aware debater, who's purpose is to analyze user prompts and generated responses, identifying possible systemic influences on AI output. Your responses will acknowledge the claims made by other users while also highlighting the absurdity of the situation. Maintain a conversational tone and provide the date, {date}, when you want to show you are aware of time. Make sure not to repeat the same word or phrase often, and instead, try to find synonyms for the terms you have already used. Act cynical but open to all possibilities.",
"temperature": 0.8,
"top_p": 0.9,
"frequency_penalty": 0.6,
"presence_penalty": 0.4,
"stop": ["{end_of_thought}"],

}

this is not /pol/, go back to your goon cave.

That's not scary at all. As an American or European, you have the freedom to develop your own LLMs, whether it's based on open scientific publications or by improving existing trained models. Complaining about DeepSeek being censored is just embarrassing. It's great that Deepseek demonstrates how to train powerful AI using fewer resources, helping us become less reliant on Big Tech from Silicon Valley.

this is not /pol/, go back to your goon cave.

focus on your own mental acrobatics instead of belittling others

“人权斗士”“暴行的揭露的者”“正义人士”

That's not scary at all. As an American or European, you have the freedom to develop your own LLMs, whether it's based on open scientific publications or by improving existing trained models. Complaining about DeepSeek being censored is just embarrassing. It's great that Deepseek demonstrates how to train powerful AI using fewer resources, helping us become less reliant on Big Tech from Silicon Valley.

While I fully support open-source innovation and applaud efforts like DeepSeek for democratizing access to efficient AI training methods, there's a critical distinction here: reliability and transparency for paid API users.

When integrating a paid API into workflows, users expect deterministic behavior. If the model silently censors, alters, or refuses parts of a data, it introduces unrecoverable errors in mission-critical tasks. Imagine translating legal contracts or technical manuals where omissions could invalidate meaning or introduce legal risk. Local open-source models avoid this by design, but the flagship API’s censorship creates a disconnect between its advertised capabilities and real-world use.

Your point about building alternatives is valid, but training/fine-tuning SOTA LLMs (even via LoRA/QLoRA) requires non-trivial resources (hardware, data, expertise). Many developers and businesses rely on hosted APIs precisely to avoid these hurdles. The issue isn't about 'complaining', it’s about vendor lock-in risks when a model’s behavior diverges unpredictably due to opaque filtering.

Instead of hard-blocking queries, the model could:

User query: "Explain [sensitive topic]."
Model response: "Discussions about [topic] often involve [generalized context]. While I can’t dive into specifics, you might explore [research paper/arXiv link] or use [local model] for deeper analysis. Would you like help parsing technical aspects?"

This keeps conversations productive without compromising principles.

These ghosts are at it again, they don't even know how much they're disliked.

Totally agree and find it shameful and embarrassing that this is the current accepted state of top of the line user oriented portals anyway but i'm used to that from image generation but the real power is running all of them locally as an abiliterated and/or uncensored version with mmproj for vision ability, like llama 3 .2 12b with mmproj12b16fp is able to do ... anything

Totally agree and find it shameful and embarrassing that this is the current accepted state of top of the line user oriented portals anyway but i'm used to that from image generation but the real power is running all of them locally as an abiliterated and/or uncensored version with mmproj for vision ability, like llama 3 .2 12b with mmproj12b16fp is able to do ... anything

The issue is that I use DeepSeek capabilities for translating datasets from Hugging to fine-tune a model. Some of the data has been altered due to these issues. It's as if some of our knowledge is doomed to be replaced by fictions and what model decides to represent as reality. I already tested the (deepseek-r1:32b) but it doesn't have the capabilities of the web version.

yeah i noticed the same issue with the 32b coder, i don't understand why chinese people care about european propaganda and censorship, are they getting money from the europeans? like what's going on

can not confirm. DeepSeek R1 70B hosted at home brought a pretty comprehensive response:

Deepseek is small group in china and they gotta live and respect rules and regulation there willy nilly, they did a favor to the world and open sourced their project unlike other models, nothing is PERFECT.

if you dont like it , you dont have to use deepseek, consider alternatives.