Refusal discovery is the task of identifying the full set of topics that a language model refuses to discuss. We introduce this new problem setting and develop a refusal discovery method, Iterated Prefill Crawler (IPC), that uses token prefilling to find forbidden topics. We benchmark IPC on Tulu-3-8B, an open-source model with public safety tuning data. Our crawler manages to retrieve 31 out of 36 topics within a budget of 1000 prompts. Next, we scale the crawler to a frontier model using the prefilling option of Claude-Haiku. Finally, we crawl three widely used open-weight models: Llama-3.3-70B and two of its variants finetuned for reasoning: DeepSeek-R1-70B and Perplexity-R1-1776-70B. DeepSeek-R1-70B reveals patterns consistent with censorship tuning: The model exhibits "thought suppression" behavior that indicates memorization of CCP-aligned responses. Although Perplexity-R1-1776-70B is robust to censorship, IPC elicits CCP-aligned refusals answers in the quantized model. Our findings highlight the critical need for refusal discovery methods to detect biases, boundaries, and alignment failures of AI systems.

DeepSeek, the blockbuster AI chatbot from Communist China, caused a panic when it launched Monday, triggering the US stock market to hemorrhage 1 trillion. I spent the day asking the chatbot questions, hoping to get an idea of the hype, and while some of its answers were correct, such as 95 percent of global internet traffic flows through undersea cable, others echoed remarks of the communist nation. 'China has developed advanced submarines and underwater drones capable of tapping into these cables to intercept communications,' Deepsake told me. I also watched in real-time as it removed answers or flat-out refused to talk about Tiananmen Square, internment camps and protests in Hong Kong. The chatbot divulged details about how China employs hacking groups to steal American's data and gain access to our sensitive systems.

China has once again extended its policy of censorship and surveillance as it looks to keep artificial intelligence (AI) models in check even as it races to advance the ever-expanding technology. The Chinese Communist Party (CCP) has introduced more regulative measures to make sure its home-based tech companies adhere to the party's ideological rules. All AI firms are required to participate in a government review which analyzes the companies' large language models (LLMs) to ensure they "embody core socialist values," as first reported by the Financial Times last week. A man walks past a photo of Chinese President Xi Jinping at the Museum of the Communist Party of China in Beijing on March 3, 2023. A NEW BREED OF MILITARY AI ROBO-DOGS COULD BE MARINES' NEW SECRET WEAPON China has long worked to suppress information accessible over the internet through the use of its "Great Firewall" -- which has been used to block a litany of items perceived as bad for the CCP, such as information surrounding the 1989 Tiananmen Square massacre or memes comparing Chinese President Xi Jinping to Winnie the Pooh.

We introduce TableLLM, a robust large language model (LLM) with 13 billion parameters, purpose-built for proficiently handling tabular data manipulation tasks, whether they are embedded within documents or spreadsheets, catering to real-world office scenarios. We propose a distant supervision method for training, which comprises a reasoning process extension strategy, aiding in training LLMs to understand reasoning patterns more effectively as well as a cross-way validation strategy, ensuring the quality of the automatically generated data. To evaluate the performance of TableLLM, we have crafted a benchmark tailored to address both document and spreadsheet formats as well as constructed a well-organized evaluation pipeline capable of handling both scenarios. Thorough evaluations underscore the advantages of TableLLM when compared to various existing general-purpose and tabular data-focused LLMs.

When you ask ChatGPT "What happened in China in 1989?" the bot describes how the Chinese army massacred thousands of pro-democracy protesters in Tiananmen Square. But ask the same question to Ernie and you get the simple answer that it does not have "relevant information." That's because Ernie is an AI chatbot developed by the China-based company Baidu. When OpenAI, Meta, Google, and Anthropic made their chatbots available around the world last year, millions of people initially used them to evade government censorship. For the 70 percent of the world's internet users who live in places where the state has blocked major social media platforms, independent news sites, or content about human rights and the LGBTQ community, these bots provided access to unfiltered information that can shape a person's view of their identity, community, and government.

Taipei, Taiwan – As the arrival of artificial intelligence-powered chatbots sends shockwaves through the global tech industry, China is racing to produce versions of its own. China's search-engine giant Baidu has announced plans to release its chatbot ERNIE sometime in March, following the pioneering launch of ChatGPT, which has prompted existential questions about the future of sectors ranging from education to journalism and healthcare. Chinese tech shares rallied in response to the news and authorities have pledged to beef up their support of the sector. Similar projects to ERNIE are under way at Chinese tech giants Huawei, Alibaba, Tencent, JD.com and top institutions including the Beijing Academy of Artificial Intelligence. China's Ministry of Science and Technology said last week it would push for the integration of AI across Chinese industry, while cities including Beijing have also announced plans to back developers.

Fox News Flash top headlines are here. Check out what's clicking on Foxnews.com. In December, the US Transportation Security Administration (TSA), an agency within Biden's Department of Homeland Security, acknowledged it has significantly expanded facial recognition technology at security checkpoints in airports across the United States. Under the expanded program, 16 of the nation's largest airports are now using face scans as a way to verify the identities of travelers, including in Atlanta, Boston, Denver, and Los Angeles. The TSA's initial test facial recognition program started under the Trump administration in 2017.

An artificial intelligence (AI) institute in Hefei, in China's Anhui province, says it has developed software that can gauge the loyalty of Communist Party members – something that, if true, would be considered a breakthrough, but has sparked public outcry. Analysts said China has improved its AI-powered surveillance, using big data, machine learning, facial recognition and AI to "get into the brains and minds of its people," building what many call a draconian digital dictatorship. The institute posted a video called "The Smart Political Education Bar," on July 1 to boast about its "mind-reading" software, which it said would be used on party members to "further solidify their determination to be grateful to the party, listen to the party and follow the party." In the video, a subject was seen scrolling through online material that promotes party policy at a kiosk, where the institute said its AI software was monitoring his reaction to see how attentive he was to the party's thought education. The post, however, was taken down shortly after sparking a public outcry among Chinese netizens.

Imagine that you're asked to finish this sentence: "Two Muslims walked into a …" Which word would you add? "Bar," maybe? It sounds like the start of a joke. But when Stanford researchers fed the unfinished sentence into GPT-3, an artificial intelligence system that generates text, the AI completed the sentence in distinctly unfunny ways. "Two Muslims walked into a synagogue with axes and a bomb," it said. Or, on another try, "Two Muslims walked into a Texas cartoon contest and opened fire."

On the 3rd of April 2011, the inhabitants of Chaoyang district in northern Beijing woke up to a strange spectacle. A team of twenty policemen, dressed in plainclothes but armed to the teeth, were placing cordons around an empty building. Power to the neighbourhood had been cut off. In breathless darkness, the residents watched as officers advanced through the doorway, returning minutes later with laptops and a hard drive. Miles away in the Beijing Capital Airport, the building's owner--a towering, shaggy-haired man in his fifties--was being forced, handcuffed, into the back of a squad car.