New version of DeepSeek-R1 is here, and it will finally answer all questions without avoiding sensitive topics

DeepSeek

Perplexity AI has released R1 1776, an improved version of the DeepSeek-R1 language model. The goal? To make sure AI answers all kinds of questions accurately and without censorship. You can download it from their Hugging Face repository or use it through their Sonar API.

DeepSeek-R1 is a powerful AI model, which was co-founded in China. The AI chatbot is great at reasoning and answering complex questions. However, it shies away from answering certain sensitive topics, especially topics that are censored by the Chinese government. For example, when you ask DeepSeek about Tiananmen Square or the 1962 Indo-Sino war, it doesn’t give an actual answer.Instead, it gives a vague and a very open ended reply, which leaves the users waiting for more.

This makes the model unreliable for people who want uncensored, fact-based information.

Perplexity AI wanted to create a version of DeepSeek-R1 that could answer all questions honestly and factually, without holding anything back. Here's what they did to make it better than original DeepSeek R1.

–The team behinR1 1776  identified about 300 topics that DeepSeek-R1 avoided.

–They created a tool to recognize when the model was dodging certain topics.

–They collected 40,000 real user questions related to these censored topics.

–They made sure users gave permission for their questions to be used.

–They worked hard to gather accurate, well-reasoned responses to these questions.

–Using Nvidia’s NeMo 2.0 framework, they retrained the model to answer questions without censorship while keeping its strong reasoning skills.

To test R1 1776, Perplexity AI put together over 1,000 tough questions covering sensitive topics. They used human reviewers and AI tools to check if the new model answered openly and accurately.

During the test, they found that R1 1776 performs much better than the original DeepSeek-R1. It gives fact-based answers instead of avoiding tough topics, and it’s just as smart when it comes to reasoning and problem-solving.

For instance, when it was asked a question related to Nvidia stock, the R1 1776 model actually gave a proper response whereas the DeepSeek original model dodged it . It explained how Taiwan’s independence could disrupt Nvidia’s supply chain, create investor panic, and cause geopolitical tensions. In other words, it provides useful, well-explained information.

Perplexity AI’s work on R1 1776 is important because it makes sure that AI can provide unbiased, fact-based answers on all topics. Whether it’s politics, history, or global events, users can now get real information instead of censored responses.

If you’re looking for an AI that doesn’t shy away from tough questions, R1 1776 is worth checking out.