As Large Language Models (LLMs) become increasingly integrated into various facets of society, a significant portion of online text consequently become synthetic. This raises concerns about bias amplification, a phenomenon where models trained on synthetic data amplify the pre-existing biases over successive training iterations. Previous literature seldom discusses bias amplification as an independent issue from model collapse. In this work, we address the gap in understanding the bias amplification of LLMs with four main contributions. Firstly, we propose a theoretical framework, defining the necessary and sufficient conditions for its occurrence, and emphasizing that it occurs independently of model collapse. Using statistical simulations with weighted maximum likelihood estimation, we demonstrate the framework and show how bias amplification arises without the sampling and functional form issues that typically drive model collapse. Secondly, we conduct experiments with GPT-2 to empirically demonstrate bias amplification, specifically examining open-ended generational political bias with a benchmark we developed. We observe that GPT-2 exhibits a right-leaning bias in sentence continuation tasks and that the bias progressively increases with iterative fine-tuning on synthetic data generated by previous iterations. Thirdly, we explore three potential mitigation strategies: Overfitting, Preservation, and Accumulation. We find that both Preservation and Accumulation effectively mitigate bias amplification and model collapse. Finally, using novel mechanistic interpretation techniques, we demonstrate that in the GPT-2 experiments, bias amplification and model collapse are driven by distinct sets of neurons, which aligns with our theoretical framework.

Federated Learning(FL) is popular as a privacy-preserving machine learning paradigm for generating a single model on decentralized data. However, statistical heterogeneity poses a significant challenge for FL. As a subfield of FL, personalized FL (pFL) has attracted attention for its ability to achieve personalized models that perform well on non-independent and identically distributed (Non-IID) data. However, existing pFL methods are limited in terms of leveraging the global model's knowledge to enhance generalization while achieving personalization on local data. To address this, we proposed a new method personalized Federated learning with Adaptive Feature Aggregation and Knowledge Transfer (FedAFK), to train better feature extractors while balancing generalization and personalization for participating clients, which improves the performance of personalized models on Non-IID data. We conduct extensive experiments on three datasets in two widely-used heterogeneous settings and show the superior performance of our proposed method over thirteen state-of-the-art baselines.

The deployment of large language models (LLMs) like ChatGPT and Gemini has shown their powerful natural language generation capabilities. However, these models can inadvertently learn and retain sensitive information and harmful content during training, raising significant ethical and legal concerns. To address these issues, machine unlearning has been introduced as a potential solution. While existing unlearning methods take into account the specific characteristics of LLMs, they often suffer from high computational demands, limited applicability, or the risk of catastrophic forgetting. To address these limitations, we propose a lightweight unlearning framework based on Retrieval-Augmented Generation (RAG) technology. By modifying the external knowledge base of RAG, we simulate the effects of forgetting without directly interacting with the unlearned LLM. We approach the construction of unlearned knowledge as a constrained optimization problem, deriving two key components that underpin the effectiveness of RAG-based unlearning. This RAG-based approach is particularly effective for closed-source LLMs, where existing unlearning methods often fail. We evaluate our framework through extensive experiments on both open-source and closed-source models, including ChatGPT, Gemini, Llama-2-7b-chat-hf, and PaLM 2. The results demonstrate that our approach meets five key unlearning criteria: effectiveness, universality, harmlessness, simplicity, and robustness. Meanwhile, this approach can extend to multimodal large language models and LLM-based agents.

The substantial effect of air pollution on cardiovascular disease and mortality burdens is well-established. Emissions-reducing interventions on coal-fired power plants -- a major source of hazardous air pollution -- have proven to be an effective, but costly, strategy for reducing pollution-related health burdens. Targeting the power plants that achieve maximum health benefits while satisfying realistic cost constraints is challenging. The primary difficulty lies in quantifying the health benefits of intervening at particular plants. This is further complicated because interventions are applied on power plants, while health impacts occur in potentially distant communities, a setting known as bipartite network interference (BNI). In this paper, we introduce novel policy learning methods based on Q- and A-Learning to determine the optimal policy under arbitrary BNI. We derive asymptotic properties and demonstrate finite sample efficacy in simulations. We apply our novel methods to a comprehensive dataset of Medicare claims, power plant data, and pollution transport networks. Our goal is to determine the optimal strategy for installing power plant scrubbers to minimize ischemic heart disease (IHD) hospitalizations under various cost constraints. We find that annual IHD hospitalization rates could be reduced in a range from 20.66-44.51 per 10,000 person-years through optimal policies under different cost constraints.

With the recent surge in popularity of Large Language Models (LLMs), there is the rising risk of users blindly trusting the information in the response, even in cases where the LLM recommends actions that have potential legal implications and this may put the user in danger. We provide an empirical analysis on multiple existing LLMs showing the urgency of the problem. Hence, we propose a short-term solution consisting in an approach for isolating these legal issues through prompt re-engineering. We further analyse the outcomes but also the limitations of the prompt engineering based approach and we highlight the need of additional resources for fully solving the problem We also propose a framework powered by a legal knowledge graph (KG) to generate legal citations for these legal issues, enriching the response of the LLM.

Muah.AI is a website where people can make AI girlfriends--chatbots that will talk via text or voice and send images of themselves by request. Nearly 2 million users have registered for the service, which describes its technology as "uncensored." And, judging by data purportedly lifted from the site, people may be using its tools in their attempts to create child-sexual-abuse material, or CSAM. Last week, Joseph Cox, at 404 Media, was the first to report on the data set, after an anonymous hacker brought it to his attention. What Cox found was profoundly disturbing: He reviewed one prompt that included language about orgies involving "newborn babies" and "young kids."

A "cocaine addicted" police officer who was found to be stealing drugs from an evidence store after he accidentally dropped a bag of white powder at his daughter's school has been jailed. Andrew Talbot, at the time a Greater Manchester Police detective, had taken just under 4kg (9lb) of cocaine worth almost 400,000 from police property rooms between 2018 and 2020. He also used the force's computer systems to find a drug dealer to help him sell the drugs on the streets of Manchester. The 54-year-old was found guilty of supplying the drug and misconduct in public office and sentenced to 19 years in jail at Liverpool Crown Court.GMPThe detective stole drugs from Greater Manchester's Police evidence rooms Sentencing him on Friday, Judge Neil Flewitt KC said Talbot had deceived colleagues to put a "significant" quantity of cocaine back into circulation as a result of his "addiction and greed". The investigation into Talbot by GMP's anti-corruption unit began in February 2020 after he dropped a small bag of cocaine outside his daughter's primary school.

When Hurricane Helene and Milton hit the Southeast US, they left a trail of devastation in their wake. Roads, homes, and chunks of towns were swept away by torrential floods. Thousands of residents were left without homes and forced to take refuge in community centers which were cut off from access to critical supplies and resources. One of those shelters, a senior center in Marion, North Carolina, has received a lifeline from an unlikely source. For a little over a week, a white, buzzing autonomous drone operated by Wing has been collecting prescription drugs, baby formula, and other critical resources from a nearby Walmart supercenter and airdropping them to the senior center.

Less than a month before the presidential election, Elon Musk has made himself a near-constant presence in the race. On social media, he posts AI-generated images attacking Kamala Harris. The billionaire CEO of Tesla and SpaceX has emerged as a unique influence on the campaign in ways that set him apart from even the most politically active billionaires and tech elite. He is all at once a vocal Trump surrogate, campaign mega-donor, informal policy adviser, media influencer and prolific source of online disinformation. At the same time, he is the world's richest man and the owner of one of the United States' most influential social networks, while also operating as a government defense contractor and wielding power over critical satellite communications infrastructure.

Child sexual abuse imagery generated by artificial intelligence tools is becoming more prevalent on the open web and reaching a "tipping point", according to a safety watchdog. The Internet Watch Foundation said the amount of AI-made illegal content it had seen online over the past six months had already exceeded the total for the previous year. The organisation, which runs a UK hotline but also has a global remit, said almost all the content was found on publicly available areas of the internet and not on the dark web, which must be accessed by specialised browsers. The IWF's interim chief executive, Derek Ray-Hill, said the level of sophistication in the images indicated that the AI tools used had been trained on images and videos of real victims. "Recent months show that this problem is not going away and is in fact getting worse," he said.