{"arxiv_id":"2608.12346","incoming":{"papers":0,"sentences":0,"groups":[],"unlabelled_sentences":0},"outgoing":{"papers":23,"sentences":29,"groups":[{"kind":"supports_this_paper","label":"Cites as support","papers":8,"sentences":8,"quotes":[{"arxiv_id":"2506.01814","title":"Analysis of LLM Bias (Chinese Propaganda & Anti-US Sentiment) in DeepSeek-R1 vs. ChatGPT o3-mini-high","url":"/paper/2506.01814","section":"How Realistic are These Threats?","sentence":"How this practically materializes can be witnessed in Chinese chatbots like DeepSeek and Baidu's Ernie Bot, which are systematically made to refuse discussing politically sensitive topics, like the 1989 Tiananmen Square massacre, while amplifying Chinese Communist Party positions, such as on Taiwan and U.S. relations [ref] .","intent_p":0.98,"contests_p":null},{"arxiv_id":"2505.12625","title":"R1dacted: Investigating Local Censorship in DeepSeek's R1 Language Model","url":"/paper/2505.12625","section":"How Realistic are These Threats?","sentence":"How this practically materializes can be witnessed in Chinese chatbots like DeepSeek and Baidu's Ernie Bot, which are systematically made to refuse discussing politically sensitive topics, like the 1989 Tiananmen Square massacre, while amplifying Chinese Communist Party positions, such as on Taiwan and U.S. relations [ref] .","intent_p":0.98,"contests_p":null},{"arxiv_id":"2306.12001","title":"An Overview of Catastrophic AI Risks","url":"/paper/2306.12001","section":"Why We Do Not Call for Stopping AI Alignment Research","sentence":"Without appropriate safeguards, criminals and even terrorists can much more easily use these models to commit crimes and harm people [ref] .","intent_p":0.99,"contests_p":null},{"arxiv_id":"2501.17805","title":"International AI Safety Report","url":"/paper/2501.17805","section":"Why We Do Not Call for Stopping AI Alignment Research","sentence":"In such a scenario, achieving the most secure and robust alignment possible transitions from important to absolutely critical [ref] .","intent_p":0.97,"contests_p":null},{"arxiv_id":"2411.06837","title":"Persuasion with Large Language Models: A Survey of Empirical Evidence, Study Methodologies, and Ethical Implications","url":"/paper/2411.06837","section":"Growing Use of AI Amplifies Misuse Impact","sentence":"Modern alignment techniques have produced systems that are highly eloquent and persuasive [ref] .","intent_p":0.96,"contests_p":null}]},{"kind":"background","label":"Cites as background","papers":1,"sentences":1,"quotes":[{"arxiv_id":"2112.00861","title":"A General Language Assistant as a Laboratory for Alignment","url":"/paper/2112.00861","section":"Post-Training Preference Alignment","sentence":"Post-training usually aims to turn a base pre-trained model into a “helpful, honest and harmless” assistant [ref] .","intent_p":0.97,"contests_p":null}]}],"unlabelled_sentences":20},"claims":[],"thresholds":{"intent_p":0.8,"supports_p":0.95,"contests_p":0.7},"method":"Citing sentences come from each paper's LaTeX source; every \\cite marks the sentence and the works it cites. Pith's classifier typed each sentence. A label is shown at intent probability 0.8 or more (0.95 for support), a contest at 0.7 or more. Measured accuracy against two blind model families is at https://pith.science/accuracy. The quoted sentences are the citing authors' own.","held_up":null,"_links":{"html":"https://pith.science/paper/2608.12346"}}