Speakers presented various perspectives on large language models (LLMs) in the context of causality and symbolic reasoning. Emre Kıcıman (Microsoft Research) emphasized that LLMs can be useful in the applied causal process, even if they don't have fully generalizable causal capabilities. Andrew Lampinen (Google DeepMind) shared the insights from his work, suggesting that LLMs can learn generalizable causal strategies under certain circumstances, but these circumstances are likely not met for the existing models. Guy van den Broeck (UCLA) presented his work on constraining and conditioning LLM generation using hidden Markov models (HMMs). Judea Pearl shared his thoughts on the possibility of LLMs learning a partial implicit world model.

How it works: Jets release heat, water vapor, and particulate matter that can produce thin clouds in the sky, known as "contrails". When numerous flights pass through such areas, these contrails can form clouds that absorb radiation escaping from the surface, acting as blankets floating above the Earth. Why it matters: A small fraction of overall flights, between 2% and 10%, create about 80% of the contrails. So the growing hope is that simply rerouting those flights could significantly reduce the effect, presenting a potentially high leverage, low cost and fast way of easing warming. The news: Since their inception, it's been clear that large language models like ChatGPT absorb racist views from the millions of pages of the internet they are trained on. Developers have responded by trying to make them less toxic.

This time, authors are suing NVIDIA over its AI platform NeMo, a language model that allows businesses to create and train their own chatbots, Ars Technica reported. They claim the company trained it on a controversial dataset that illegally used their books without consent. Authors Abdi Nazemian, Brian Keene and Stewart O'Nan demanded a jury trial and asked Nvidia to pay damages and destroy all copies of the Books3 dataset used to power NeMo large language models (LLMs). They claim that dataset copied a shadow library called Bibliotek consisting of 196,640 pirated books. "In sum, NVIDIA has admitted training its NeMo Megatron models on a copy of The Pile dataset," the claim states.

"There is no Founding Agreement, or any agreement at all with Musk," OpenAI said in a court filing as a defendant in Elon Musk's lawsuit. We're, of course, talking about the lawsuit Musk filed against OpenAI, which accuses it of violating its status as a non-profit, as well as of violating a founding agreement promising the organization would never operate for profit and would release its AI publicly. The company said the billionaire's claims are based on "convoluted -- often incoherent -- factual premises." It called that founding agreement "a fiction Musk has conjured to lay unearned claim to the fruits of an enterprise he initially supported, then abandoned, then watched succeed without him." If the case goes to discovery, there's evidence that would show that Musk supported OpenAI's transition into a for-profit structure, "to be controlled by Musk himself," OpenAI continued.

Personalized dialogue systems have gained significant attention in recent years for their ability to generate responses in alignment with different personas. However, most existing approaches rely on pre-defined personal profiles, which are not only time-consuming and labor-intensive to create but also lack flexibility. We propose In-Dialogue Learning (IDL), a fine-tuning framework that enhances the ability of pre-trained large language models to leverage dialogue history to characterize persona for completing personalized dialogue generation tasks without pre-defined profiles. Our experiments on three datasets demonstrate that IDL brings substantial improvements, with BLEU and ROUGE scores increasing by up to 200% and 247%, respectively. Additionally, the results of human evaluations further validate the efficacy of our proposed method.

Large language models are playing an increasingly significant role in molecular research, yet existing models often generate erroneous information, posing challenges to accurate molecular comprehension. Traditional evaluation metrics for generated content fail to assess a model's accuracy in molecular understanding. To rectify the absence of factual evaluation, we present MoleculeQA, a novel question answering (QA) dataset which possesses 62K QA pairs over 23K molecules. Each QA pair, composed of a manual question, a positive option and three negative options, has consistent semantics with a molecular description from authoritative molecular corpus. MoleculeQA is not only the first benchmark for molecular factual bias evaluation but also the largest QA dataset for molecular research. A comprehensive evaluation on MoleculeQA for existing molecular LLMs exposes their deficiencies in specific areas and pinpoints several particularly crucial factors for molecular understanding.

We study how to watermark LLM outputs, i.e. embedding algorithmically detectable signals into LLM-generated text to track misuse. Unlike the current mainstream methods that work with a fixed LLM, we expand the watermark design space by including the LLM tuning stage in the watermark pipeline. While prior works focus on token-level watermark that embeds signals into the output, we design a model-level watermark that embeds signals into the LLM weights, and such signals can be detected by a paired detector. We propose a co-training framework based on reinforcement learning that iteratively (1) trains a detector to detect the generated watermarked text and (2) tunes the LLM to generate text easily detectable by the detector while keeping its normal utility. We empirically show that our watermarks are more accurate, robust, and adaptable (to new attacks). It also allows watermarked model open-sourcing. In addition, if used together with alignment, the extra overhead introduced is low - only training an extra reward model (i.e. our detector). We hope our work can bring more effort into studying a broader watermark design that is not limited to working with a fixed LLM. We open-source the code: https://github.com/xiaojunxu/learning-to-watermark-llm .

To thoroughly assess the mathematical reasoning abilities of Large Language Models (LLMs), we need to carefully curate evaluation datasets covering diverse mathematical concepts and mathematical problems at different difficulty levels. In pursuit of this objective, we propose FineMath in this paper, a fine-grained mathematical evaluation benchmark dataset for assessing Chinese LLMs. FineMath is created to cover the major key mathematical concepts taught in elementary school math, which are further divided into 17 categories of math word problems, enabling in-depth analysis of mathematical reasoning abilities of LLMs. All the 17 categories of math word problems are manually annotated with their difficulty levels according to the number of reasoning steps required to solve these problems. We conduct extensive experiments on a wide range of LLMs on FineMath and find that there is still considerable room for improvements in terms of mathematical reasoning capability of Chinese LLMs. We also carry out an in-depth analysis on the evaluation process and methods that have been overlooked previously. These two factors significantly influence the model results and our understanding of their mathematical reasoning capabilities. The dataset will be publicly available soon.

Large language models (LLMs) are widely deployed in various downstream tasks, e.g., auto-completion, aided writing, or chat-based text generation. However, the considered output candidates of the underlying search algorithm are under-explored and under-explained. We tackle this shortcoming by proposing a tree-in-the-loop approach, where a visual representation of the beam search tree is the central component for analyzing, explaining, and adapting the generated outputs. To support these tasks, we present generAItor, a visual analytics technique, augmenting the central beam search tree with various task-specific widgets, providing targeted visualizations and interaction possibilities. Our approach allows interactions on multiple levels and offers an iterative pipeline that encompasses generating, exploring, and comparing output candidates, as well as fine-tuning the model based on adapted data. Our case study shows that our tool generates new insights in gender bias analysis beyond state-of-the-art template-based methods. Additionally, we demonstrate the applicability of our approach in a qualitative user study. Finally, we quantitatively evaluate the adaptability of the model to few samples, as occurring in text-generation use cases.

Recent advancements in Natural Language Processing (NLP) with Large Language Models (LLMs) have made the sentiment analysis of financial news by machines a practical achievement and no longer just a dream. More precisely, Large Language Models (LLMs) have marked a major step forward in processing large contexts, exhibiting human-level performance on various professional and academic benchmarks, although they still have limitations such as reliability issues and limited context windows [OpenAI, 2023]. Their ability to process more context has shown particularly interesting applications in many business areas [George and George, 2023]. Hence exploring the potential to extract either weak or strong signals from financial news to enhance a risk-on risk-off investment strategy becomes highly pertinent. Indeed, extracting sentiment from financial news is not new [Tetlock, 2007, Schumaker and Chen, 2009], and finance has a longstanding tradition of exploiting textual data [Kearney and Liu, 2014].