DeeperML

Perplexity AI has launched "R1 1776," a modified version of the open-source language model DeepSeek R1, effectively stripping away its built-in Chinese censorship. The original DeepSeek R1, developed in China, gained recognition for its reasoning capabilities, but responses to queries on sensitive topics such as Chinese history and geopolitics were often censored or aligned with pro-government stances. Perplexity AI's modified model, dubbed R1 1776 (evoking the spirit of independence), aims to address these limitations.

The team at Perplexity AI identified approximately 300 sensitive topics that were subject to censorship in the original DeepSeek R1. They then curated a dataset of prompts designed to elicit censored responses and, using post-training techniques, re-trained the model to provide more open-ended and contextually accurate answers. According to Perplexity AI's testing, R1 1776 comprehensively addresses previously censored topics without bias, and its core reasoning capabilities remain unchanged. The modified model is now available on the Sonar AI platform, with model weights publicly hosted on GitHub.

ImgSrc: assets.eweek.co

References :

• the-decoder.com: Perplexity AI removes Chinese censorship from Deepseek R1
• www.unite.ai: Perplexity AI “Uncensorsâ€� DeepSeek R1: Who Decides AI’s Boundaries?
• www.eweek.com: The modifications change the model’s responses to Chinese history and geopolitics prompts. DeepSeek-R1 is open source.
• eWEEK: Perplexity 1776 Model Fixes DeepSeek-R1’s “Refusal to Respond to Sensitive Topicsâ€�
• Unite.AI: Perplexity AI “Uncensorsâ€� DeepSeek R1: Who Decides AI’s Boundaries?
• www.producthunt.com: R1 1776

Classification:

• HashTags: #AIethics #Censorship #DeepSeek
• Company: Perplexity AI
• Product: DeepSeek
• Feature: Uncensored
• Malware: R1 1776
• Type: AI
• Severity: Informative