aback
Adaptive Backtracking for Privacy Protection in Large Language Models
Yao, Zhihao, Gu, Yuxuan, Feng, Xiachong, Ma, Weitao, Li, Bo, Feng, Xiaocheng
The preservation of privacy has emerged as a critical topic in the era of artificial intelligence. However, current work focuses on user-oriented privacy, overlooking severe enterprise data leakage risks exacerbated by the Retrieval-Augmented Generation paradigm. To address this gap, our paper introduces a novel objective: enterprise-oriented privacy concerns. Achieving this objective requires overcoming two fundamental challenges: existing methods such as data sanitization severely degrade model performance, and the field lacks public datasets for evaluation. We address these challenges with several solutions. (1) To prevent performance degradation, we propose ABack, a training-free mechanism that leverages a Hidden State Model to pinpoint the origin of a leakage intention and rewrite the output safely. (2) To solve the lack of datasets, we construct PriGenQA, a new benchmark for enterprise privacy scenarios in healthcare and finance. To ensure a rigorous evaluation, we move beyond simple static attacks by developing a powerful adaptive attacker with Group Relative Policy Optimization. Experiments show that against this superior adversary, ABack improves the overall privacy utility score by up to 15\% over strong baselines, avoiding the performance trade-offs of prior methods.
OpenAI staffers reportedly 'taken aback' by 'ominous' logo rebranding
OpenAI could undergo massive changes next year, which include getting a brand new logo. According to Fortune, though, staff members were less than enthused when they got a sneak peek of its supposed new logo at a recent company-wide meeting. The company's hexagonal flower symbol, which has become pretty recognizable thanks to ChatGPT's popularity, is gone. Instead, it's replaced by a large black "O" or a simple ring or circle that staffers reportedly found to be devoid of creativity -- ominous, even. Based on how the publication's sources described it, the new logo sounds like the complete opposite of OpenAI's current one, which was designed to represent "precision, potential and optimism."
Kamala Harris taken aback by CBS host asking about Trump's re-election hopes: 'Don't understand the question'
Vice President Harris appeared stunned by CBS' Margaret Brennan's question about whether she was taking the threat of another Trump presidency "seriously enough." Vice President Kamala Harris appeared stunned by a question from CBS host Margaret Brennan, who wondered if she was taking the possibility of another Donald Trump presidency "seriously enough." Harris looked taken aback, pausing before responding, "I don't understand the question." "You were dismissive of some of the Republican criticism of you and the president. When you look at current polling, the frontrunner for the Republican nomination is the former president, the 45th president," Brennan added on "Face The Nation."
Banwen rave: Eight fined and arrests made for drug driving
Eight people have now been fined up to £10,000 after an illegal rave that attracted 3,000 people, with arrests also made for public order offences and driving under the influence of drugs. The unlicensed event at Banwen, on the edge of the Brecon Beacons, started Saturday night. There were still 400 people at the site on Monday morning. South Wales Police Assistant Chief Constable Dave Thorne said drone footage would help identify organisers. A student who attended the rave admitted being taken aback by the scale of the event and likened it to a festival.