Yesterday afternoon, Elon Musk fired the latest shot in his feud with OpenAI: His new AI venture, xAI, now allows anyone to download and use the computer code for its flagship software. No fees, no restrictions, just Grok, a large language model that Musk has positioned against OpenAI's GPT-4, the model powering the most advanced version of ChatGPT. Sharing Grok's code is a thinly veiled provocation. Musk was one of OpenAI's original backers. He left in 2018 and recently sued for breach of contract, arguing that the start-up and its CEO, Sam Altman, have betrayed the organization's founding principles in pursuit of profit, transforming a utopian vision of technology that "benefits all of humanity" into yet another opaque corporation.

Apple is reportedly in talks with Google to integrate its Gemini AI in iPhones, Bloomberg reports, a move that should help both companies compete with OpenAI and its (heavily invested) partner Microsoft. While it might seem like an admission that Apple is lagging behind on AI, the partnership fits if you think of generative AI models as an evolution of web searching, something Google already provides to all of Apple's devices. According to the report, Gemini could be the cloud-based generative AI engine for Siri and other iPhone apps, while Apple's models could be woven into the upcoming iOS 18 for on-device AI tasks. Bloomberg notes that Apple has also had discussions with OpenAI about using its own models, and it could still end up partnering with another AI outfit, like Anthropic. Apple could conceivably even work with multiple partners until its own generative models are up to snuff.

Accepted papers spanned a diverse range of topics in cutting edge AI research and applications. This included computer vision, multimodal learning, fairness and ethics considerations, interpretability and explainability of models, learning effective representations, continual learning, generative modeling techniques, and novel applications in healthcare among others. We gave awards to three papers which share the common goal of aligning AI models, especially large language models, with human values, preferences and social intelligence. One proposes techniques for improved controllability of language model outputs through activation steering, allowing humans to guide model behavior. Another explores hybrid natural language and feedback signals to fine-tune models towards satisfying human feedback during training itself.

Extracting semantic information from generated text is a useful tool for applications such as automated fact checking or retrieval augmented generation. Currently, this requires either separate models during inference, which increases computational cost, or destructive fine-tuning of the language model. Instead, we propose directly embedding information extraction capabilities into pre-trained language models using probing classifiers, enabling efficient simultaneous text generation and information extraction. For this, we introduce an approach called EMBER and show that it enables named entity recognition in decoder-only language models without fine-tuning them and while incurring minimal additional computational cost at inference time. Specifically, our experiments using GPT-2 show that EMBER maintains high token generation rates during streaming text generation, with only a negligible decrease in speed of around 1% compared to a 43.64% slowdown measured for a baseline using a separate NER model. Code and data are available at https://github.com/nicpopovic/EMBER.

Language models (LMs) are statistical models trained to assign probability to human-generated text. As such, it is reasonable to question whether they approximate linguistic variability exhibited by humans well. This form of statistical assessment is difficult to perform at the passage level, for it requires acceptability judgements (i.e., human evaluation) or a robust automated proxy (which is non-trivial). At the word level, however, given some context, samples from an LM can be assessed via exact matching against a prerecorded dataset of alternative single-word continuations of the available context. We exploit this fact and evaluate the LM's ability to reproduce variability that humans (in particular, a population of English speakers) exhibit in the 'next word prediction' task. This can be seen as assessing a form of calibration, which, in the context of text classification, Baan et al. (2022) termed calibration to human uncertainty. We assess GPT2, BLOOM and ChatGPT and find that they exhibit fairly low calibration to human uncertainty. We also verify the failure of expected calibration error (ECE) to reflect this, and as such, advise the community against relying on it in this setting.

Advanced life forms, sustained by the synergistic interaction of neural cognitive mechanisms, continually acquire and transfer knowledge throughout their lifespan. In contrast, contemporary machine learning paradigms exhibit limitations in emulating the facets of continual learning (CL). Nonetheless, the emergence of large language models (LLMs) presents promising avenues for realizing CL via interactions with these models. Drawing on Complementary Learning System theory, this paper presents a novel Interactive Continual Learning (ICL) framework, enabled by collaborative interactions among models of various sizes. Specifically, we assign the ViT model as System1 and multimodal LLM as System2. To enable the memory module to deduce tasks from class information and enhance Set2Set retrieval, we propose the Class-Knowledge-Task Multi-Head Attention (CKT-MHA). Additionally, to improve memory retrieval in System1 through enhanced geometric representation, we introduce the CL-vMF mechanism, based on the von Mises-Fisher (vMF) distribution. Meanwhile, we introduce the von Mises-Fisher Outlier Detection and Interaction (vMF-ODI) strategy to identify hard examples, thus enhancing collaboration between System1 and System2 for complex reasoning realization. Comprehensive evaluation of our proposed ICL demonstrates significant resistance to forgetting and superior performance relative to existing methods. Code is available at github.com/ICL.

Large Language Models (LLMs) are composed of neurons that exhibit various behaviors and roles, which become increasingly diversified as models scale. Recent studies have revealed that not all neurons are active across different datasets, and this sparsity correlates positively with the task-specific ability, leading to advancements in model pruning and training efficiency. Traditional fine-tuning methods engage all parameters of LLMs, which is computationally expensive and may not be necessary. In contrast, Parameter-Efficient Fine-Tuning (PEFT) approaches aim to minimize the number of trainable parameters, yet they still operate at a relatively macro scale (e.g., layer-level). We introduce Neuron-Level Fine-Tuning (NeFT), a novel approach that refines the granularity of parameter training down to the individual neuron, enabling more precise and computationally efficient model updates. The experimental results show that NeFT not only exceeded the performance of full-parameter fine-tuning and PEFT but also provided insights into the analysis of neurons.

This research paper investigates public views on climate change and biodiversity loss by analyzing questions asked to the ClimateQ&A platform. ClimateQ&A is a conversational agent that uses LLMs to respond to queries based on over 14,000 pages of scientific literature from the IPCC and IPBES reports. Launched online in March 2023, the tool has gathered over 30,000 questions, mainly from a French audience. Its chatbot interface allows for the free formulation of questions related to nature*. While its main goal is to make nature science more accessible, it also allows for the collection and analysis of questions and their themes. Unlike traditional surveys involving closed questions, this novel method offers a fresh perspective on individual interrogations about nature. Running NLP clustering algorithms on a sample of 3,425 questions, we find that a significant 25.8% inquire about how climate change and biodiversity loss will affect them personally (e.g., where they live or vacation, their consumption habits) and the specific impacts of their actions on nature (e.g., transportation or food choices). This suggests that traditional methods of surveying may not identify all existing knowledge gaps, and that relying solely on IPCC and IPBES reports may not address all individual inquiries about climate and biodiversity, potentially affecting public understanding and action on these issues. *we use 'nature' as an umbrella term for 'climate change' and 'biodiversity loss'

In the field of medical image analysis, the scarcity of Chinese chest X-ray report datasets has hindered the development of technology for generating Chinese chest X-ray reports. On one hand, the construction of a Chinese chest X-ray report dataset is limited by the time-consuming and costly process of accurate expert disease annotation. On the other hand, a single natural language generation metric is commonly used to evaluate the similarity between generated and ground-truth reports, while the clinical accuracy and effectiveness of the generated reports rely on an accurate disease labeler (classifier). To address the issues, this study proposes a disease labeler tailored for the generation of Chinese chest X-ray reports. This labeler leverages a dual BERT architecture to handle diagnostic reports and clinical information separately and constructs a hierarchical label learning algorithm based on the affiliation between diseases and body parts to enhance text classification performance. Utilizing this disease labeler, a Chinese chest X-ray report dataset comprising 51,262 report samples was established. Finally, experiments and analyses were conducted on a subset of expert-annotated Chinese chest X-ray reports, validating the effectiveness of the proposed disease labeler.

Dataspaces have recently gained adoption across various sectors, including traditionally less digitized domains such as culture. Leveraging Semantic Web technologies helps to make dataspaces FAIR, but their complexity poses a significant challenge to the adoption of dataspaces and increases their cost. The advent of Large Language Models (LLMs) raises the question of how these models can support the adoption of FAIR dataspaces. In this work, we demonstrate the potential of LLMs in dataspaces with a concrete example. We also derive a research agenda for exploring this emerging field.