The Center for Investigative Reporting (CIR), which publishes Mother Jones and Reveal, said on Thursday that it had filed the lawsuit accusing the tech firms of using its content without permission in a "rebuke to artificial intelligence and its exploitative practices". "OpenAI and Microsoft started vacuuming up our stories to make their product more powerful, but they never asked for permission or offered compensation, unlike other organisations that license our material," Monika Bauerlein, CEO of the Center for Investigative Reporting, said in a statement. The work of journalists, at CIR and everywhere, is valuable, and OpenAI and Microsoft know it." OpenAI and Microsoft did not immediately respond to requests for comment. OpenAI's ChatGPT chatbot relies on vast quantities of information scraped from the internet, including news sites, to respond to users' queries.

Aligning language models (LMs) to human preferences has emerged as a critical pursuit, enabling these models to better serve diverse user needs. Existing methods primarily focus on optimizing LMs for a single reward function, limiting their adaptability to varied objectives. Here, we propose $\textbf{multi-objective decoding (MOD)}$, a decoding-time algorithm that outputs the next token from a linear combination of predictions of all base models, for any given weightings over different objectives. We exploit a common form among a family of $f$-divergence regularized alignment approaches (such as PPO, DPO, and their variants) to identify a closed-form solution by Legendre transform, and derive an efficient decoding strategy. Theoretically, we show why existing approaches can be sub-optimal even in natural settings and obtain optimality guarantees for our method. Empirical results demonstrate the effectiveness of the algorithm. For example, compared to a parameter-merging baseline, MOD achieves 12.8% overall reward improvement when equally optimizing towards $3$ objectives. Moreover, we experiment with MOD on combining three fully-finetuned LLMs of different model sizes, each aimed at different objectives such as safety, coding, and general user preference. Unlike traditional methods that require careful curation of a mixture of datasets to achieve comprehensive improvement, we can quickly experiment with preference weightings using MOD to find the best combination of models. Our best combination reduces toxicity on Toxigen to nearly 0% and achieves 7.9--33.3% improvement across other three metrics ($\textit{i.e.}$, Codex@1, GSM-COT, BBH-COT).

Human and model-generated texts can be distinguished by examining the magnitude of likelihood in language. However, it is becoming increasingly difficult as language model's capabilities of generating human-like texts keep evolving. This study provides a new perspective by using the relative likelihood values instead of absolute ones, and extracting useful features from the spectrum-view of likelihood for the human-model text detection task. We propose a detection procedure with two classification methods, supervised and heuristic-based, respectively, which results in competitive performances with previous zero-shot detection methods and a new state-of-the-art on short-text detection. Our method can also reveal subtle differences between human and model languages, which find theoretical roots in psycholinguistics studies. Our code is available at https://github.com/CLCS-SUSTech/FourierGPT

Black-box finetuning is an emerging interface for adapting state-of-the-art language models to user needs. However, such access may also let malicious actors undermine model safety. To demonstrate the challenge of defending finetuning interfaces, we introduce covert malicious finetuning, a method to compromise model safety via finetuning while evading detection. Our method constructs a malicious dataset where every individual datapoint appears innocuous, but finetuning on the dataset teaches the model to respond to encoded harmful requests with encoded harmful responses. Applied to GPT-4, our method produces a finetuned model that acts on harmful instructions 99% of the time and avoids detection by defense mechanisms such as dataset inspection, safety evaluations, and input/output classifiers. Our findings question whether black-box finetuning access can be secured against sophisticated adversaries.

Large language models (LLMs) have gained significant attention due to their ability to mimic human language. Identifying texts generated by LLMs is crucial for understanding their capabilities and mitigating potential consequences. This paper analyzes datasets of varying text lengths: small, medium, and large. We compare the performance of machine learning algorithms on four datasets: (1) small (tweets from Election, FIFA, and Game of Thrones), (2) medium (Wikipedia introductions and PubMed abstracts), and (3) large (OpenAI web text dataset). Our results indicate that LLMs with very large parameters (such as the XL-1542 variant of GPT2 with 1542 million parameters) were harder (74%) to detect using traditional machine learning methods. However, detecting texts of varying lengths from LLMs with smaller parameters (762 million or less) can be done with high accuracy (96% and above). We examine the characteristics of human and machine-generated texts across multiple dimensions, including linguistics, personality, sentiment, bias, and morality. Our findings indicate that machine-generated texts generally have higher readability and closely mimic human moral judgments but differ in personality traits. SVM and Voting Classifier (VC) models consistently achieve high performance across most datasets, while Decision Tree (DT) models show the lowest performance. Model performance drops when dealing with rephrased texts, particularly shorter texts like tweets. This study underscores the challenges and importance of detecting LLM-generated texts and suggests directions for future research to improve detection methods and understand the nuanced capabilities of LLMs.

From organizing recorded videos and meetings into chapters, to breaking down large inputs in order to fit them into the context window of commoditized Large Language Models (LLMs), topic segmentation of large transcripts emerges as a task of increasing significance. Still, accurate segmentation presents many challenges, including (a) the noisy nature of the Automatic Speech Recognition (ASR) software typically used to obtain the transcripts, (b) the lack of diverse labeled data and (c) the difficulty in pin-pointing the ground-truth number of segments. In this work we present TreeSeg, an approach that combines off-the-shelf embedding models with divisive clustering, to generate hierarchical, structured segmentations of transcripts in the form of binary trees. Our approach is robust to noise and can handle large transcripts efficiently. We evaluate TreeSeg on the ICSI and AMI corpora, demonstrating that it outperforms all baselines. Finally, we introduce TinyRec, a small-scale corpus of manually annotated transcripts, obtained from self-recorded video sessions.

In the dawn of the age of Large Language Models (LLMs), already much has been said about how researchers are making use of LLMs to author articles. For example, according to an article in Scientific American [1], "One percent of scientific articles published in 2023 showed signs of generative AI's potential involvement, according to a recent analysis." However, far less has been said about how reviewers are now abusing their role, sometimes with the editor's collusion. Here is our report of a case in point. We submitted a manuscript on domain-independent deception detection to a highly respected journal. As a consequence of a reviewer's use of an LLM, we both received a most peculiar review and also lost the promised confidentiality regarding our submission.

Maddison seen progress in two popular directions - program induction and Tarlow (2014) improved this approach by introducing a and program synthesis. Program induction is the process of state vector to condition child node expansion, enhancing automatically generating computer programs from a set of the generative process. This idea was later adapted by input-output examples or specifications. "Learning to Allamanis et al. (2015) for text-to-code retrieval, Execute" by Wojciech Zaremba and Ilya Sutskever (2014) demonstrating its effectiveness in matching code snippets to explores the capabilities of Long Short-Term Memory textual queries. Yin and Neubig (2017) further extended this (LSTM) networks in learning to execute simple programs by concept to text-conditional code generation, enabling the treating the problem as a sequence-to-sequence task.

Smart home automation systems aim to improve the comfort and convenience of users in their living environment. However, adapting automation to user needs remains a challenge. Indeed, many systems still rely on hand-crafted routines for each smart object.This paper presents an original smart home architecture leveraging Large Language Models (LLMs) and user preferences to push the boundaries of personalisation and intuitiveness in the home environment.This article explores a human-centred approach that uses the general knowledge provided by LLMs to learn and facilitate interactions with the environment.The advantages of the proposed model are demonstrated on a set of scenarios, as well as a comparative analysis with various LLM implementations. Some metrics are assessed to determine the system's ability to maintain comfort, safety, and user preferences. The paper details the approach to real-world implementation and evaluation.The proposed approach of using preferences shows up to 52.3% increase in average grade, and with an average processing time reduced by 35.6% on Starling 7B Alpha LLM. In addition, performance is 26.4% better than the results of the larger models without preferences, with processing time almost 20 times faster.

Despite the vast repository of global medical knowledge predominantly being in English, local languages are crucial for delivering tailored healthcare services, particularly in areas with limited medical resources. To extend the reach of medical AI advancements to a broader population, we aim to develop medical LLMs across the six most widely spoken languages, encompassing a global population of 6.1 billion. This effort culminates in the creation of the ApolloCorpora multilingual medical dataset and the XMedBench benchmark. In the multilingual medical benchmark, the released Apollo models, at various relatively-small sizes (i.e., 0.5B, 1.8B, 2B, 6B, and 7B), achieve the best performance among models of equivalent size. Especially, Apollo-7B is the state-of-the-art multilingual medical LLMs up to 70B. Additionally, these lite models could be used to improve the multi-lingual medical capabilities of larger models without fine-tuning in a proxy-tuning fashion.