Despite showing increasingly human-like abilities, large language models (LLMs) often struggle with factual inaccuracies, i.e. "hallucinations", even when they hold relevant knowledge. To address these hallucinations, current approaches typically necessitate high-quality human factuality annotations. In this work, we explore Self-Alignment for Factuality, where we leverage the self-evaluation capability of an LLM to provide training signals that steer the model towards factuality. Specifically, we incorporate Self-Eval, a self-evaluation component, to prompt an LLM to validate the factuality of its own generated responses solely based on its internal knowledge. Additionally, we design Self-Knowledge Tuning (SK-Tuning) to augment the LLM's self-evaluation ability by improving the model's confidence estimation and calibration. We then utilize these self-annotated responses to fine-tune the model via Direct Preference Optimization algorithm. We show that the proposed self-alignment approach substantially enhances factual accuracy over Llama family models across three key knowledge-intensive tasks on TruthfulQA and BioGEN.

Large Language Models (LLMs) are susceptible to Jailbreaking attacks, which aim to extract harmful information by subtly modifying the attack query. As defense mechanisms evolve, directly obtaining harmful information becomes increasingly challenging for Jailbreaking attacks. In this work, inspired by human practices of indirect context to elicit harmful information, we focus on a new attack form called Contextual Interaction Attack. The idea relies on the autoregressive nature of the generation process in LLMs. We contend that the prior context--the information preceding the attack query--plays a pivotal role in enabling potent Jailbreaking attacks. Specifically, we propose an approach that leverages preliminary question-answer pairs to interact with the LLM. By doing so, we guide the responses of the model toward revealing the 'desired' harmful information. We conduct experiments on four different LLMs and demonstrate the efficacy of this attack, which is black-box and can also transfer across LLMs. We believe this can lead to further developments and understanding of the context vector in LLMs.

Lyft beat estimates for fourth-quarter profit on Tuesday and said it would generate positive free cash flow for the first time in 2024, as the ride-share platform reaps the benefits of heavy cost cuts. Company shares surged nearly 60% in extended trading but erased a third of those gains after the CFO corrected a major mistake in the earnings report. Erin Brewer had said that the company would grow by 500 basis points (5%) in 2024, but later said that the real increase would be a factor of 10 lower – 50 basis points (0.5%). In 2023, the stock gained about 36%. Rides to stadiums grew more than 35% last year from 2022, mainly driven by Taylor Swift's Eras Tour, Beyoncé's Renaissance World Tour and sporting events, Lyft said.

An Israeli drone strike has targeted two journalists in Muraj, north of Rafah, including Al Jazeera Arabic correspondent, Ismail Abu Omar who doctors say is in a critical condition.

Two journalists, including an Al Jazeera reporter, have been wounded in an Israeli attack north of Rafah in southern Gaza. The condition of Al Jazeera Arabic correspondent Ismail Abu Omar and his cameraman Ahmad Matar was described as serious and both were transferred to the European Gaza Hospital in Khan Younis for treatment on Tuesday. Abu Omar has had his right leg amputated, but pieces of shrapnel remained in his head and chest. Doctors were trying to save his left leg. He was undergoing surgery after suffering significant blood loss from a possible cut in the femoral artery.

Reward finetuning has emerged as a promising approach to aligning foundation models with downstream objectives. Remarkable success has been achieved in the language domain by using reinforcement learning (RL) to maximize rewards that reflect human preference. However, in the vision domain, existing RL-based reward finetuning methods are limited by their instability in large-scale training, rendering them incapable of generalizing to complex, unseen prompts. In this paper, we propose Proximal Reward Difference Prediction (PRDP), enabling stable black-box reward finetuning for diffusion models for the first time on large-scale prompt datasets with over 100K prompts. Our key innovation is the Reward Difference Prediction (RDP) objective that has the same optimal solution as the RL objective while enjoying better training stability. Specifically, the RDP objective is a supervised regression objective that tasks the diffusion model with predicting the reward difference of generated image pairs from their denoising trajectories. We theoretically prove that the diffusion model that obtains perfect reward difference prediction is exactly the maximizer of the RL objective. We further develop an online algorithm with proximal updates to stably optimize the RDP objective. In experiments, we demonstrate that PRDP can match the reward maximization ability of well-established RL-based methods in small-scale training. Furthermore, through large-scale training on text prompts from the Human Preference Dataset v2 and the Pick-a-Pic v1 dataset, PRDP achieves superior generation quality on a diverse set of complex, unseen prompts whereas RL-based methods completely fail.

Do current large language models (LLMs) better solve graph reasoning and generation tasks with parameter updates? In this paper, we propose InstructGraph, a framework that empowers LLMs with the abilities of graph reasoning and generation by instruction tuning and preference alignment. Specifically, we first propose a structured format verbalizer to unify all graph data into a universal code-like format, which can simply represent the graph without any external graph-specific encoders. Furthermore, a graph instruction tuning stage is introduced to guide LLMs in solving graph reasoning and generation tasks. Finally, we identify potential hallucination problems in graph tasks and sample negative instances for preference alignment, the target of which is to enhance the output's reliability of the model. Extensive experiments across multiple graph-centric tasks exhibit that InstructGraph can achieve the best performance and outperform GPT-4 and LLaMA2 by more than 13\% and 38\%, respectively.

Incremental learning (IL) is essential to realize the human-level intelligence in the neural network. However, existing IL scenarios and datasets are unqualified for assessing forgetting in PLMs, giving an illusion that PLMs do not suffer from catastrophic forgetting. To this end, we propose a challenging IL scenario called instance-incremental learning (IIL) and a novel dataset called Concept-1K, which supports an order of magnitude larger IL steps. Based on the experiments on Concept-1K, we reveal that billion-parameter PLMs still suffer from catastrophic forgetting, and the forgetting is affected by both model scale, pretraining, and buffer size. Furthermore, existing IL methods and a popular finetuning technique, LoRA, fail to achieve satisfactory performance. Our study provides a novel scenario for future studies to explore the catastrophic forgetting of PLMs and encourage more powerful techniques to be designed for alleviating the forgetting in PLMs. The data, code and scripts are publicly available at https://github.com/zzz47zzz/pretrained-lm-for-incremental-learning.

As Large Language Models (LLMs) become more proficient, their misuse in large-scale viral disinformation campaigns is a growing concern. This study explores the capability of ChatGPT to generate unconditioned claims about the war in Ukraine, an event beyond its knowledge cutoff, and evaluates whether such claims can be differentiated by human readers and automated tools from human-written ones. We compare war-related claims from ClaimReview, authored by IFCN-registered fact-checkers, and similar short-form content generated by ChatGPT. We demonstrate that ChatGPT can produce realistic, target-specific disinformation cheaply, fast, and at scale, and that these claims cannot be reliably distinguished by humans or existing automated tools.

In the era of widespread social networks, the rapid dissemination of fake news has emerged as a significant threat, inflicting detrimental consequences across various dimensions of people's lives. Machine learning and deep learning approaches have been extensively employed for identifying fake news. However, a significant challenge in identifying fake news is the limited availability of labeled news datasets. Therefore, the One-Class Learning (OCL) approach, utilizing only a small set of labeled data from the interest class, can be a suitable approach to address this challenge. On the other hand, representing data as a graph enables access to diverse content and structural information, and label propagation methods on graphs can be effective in predicting node labels. In this paper, we adopt a graph-based model for data representation and introduce a semi-supervised and one-class approach for fake news detection, called LOSS-GAT. Initially, we employ a two-step label propagation algorithm, utilizing Graph Neural Networks (GNNs) as an initial classifier to categorize news into two groups: interest (fake) and non-interest (real). Subsequently, we enhance the graph structure using structural augmentation techniques. Ultimately, we predict the final labels for all unlabeled data using a GNN that induces randomness within the local neighborhood of nodes through the aggregation function. We evaluate our proposed method on five common datasets and compare the results against a set of baseline models, including both OCL and binary labeled models. The results demonstrate that LOSS-GAT achieves a notable improvement, surpassing 10%, with the advantage of utilizing only a limited set of labeled fake news. Noteworthy, LOSS-GAT even outperforms binary labeled models.