The European Union's Artificial Intelligence (AI) Act is set to be a landmark legal instrument for regulating AI technology. While stakeholders have primarily focused on the governance of fixed purpose AI applications (also known as narrow AI), more attention is required to understand the nature of highly and broadly capable systems. As of the beginning of 2023, several definitions for General Purpose AI Systems (GPAIS) exist in relation to the AI Act, attempting to distinguish between systems with and without a fixed purpose. In this article, we operationalise these differences through the concept of "distinct tasks" and examine four approaches (quantity, performance, adaptability, and emergence) to determine whether an AI system should be classified as a GPAIS. We suggest that EU stakeholders use the four approaches as a starting point to discriminate between fixed-purpose and GPAIS.

Recent research has focused on enhancing the capability of smaller models through imitation learning, drawing on the outputs generated by large foundation models (LFMs). A number of issues impact the quality of these models, ranging from limited imitation signals from shallow LFM outputs; small scale homogeneous training data; and most notably a lack of rigorous evaluation resulting in overestimating the small model's capability as they tend to learn to imitate the style, but not the reasoning process of LFMs. To address these challenges, we develop Orca (We are working with our legal team to publicly release a diff of the model weights in accordance with LLaMA's release policy to be published at https://aka.ms/orca-lm), a 13-billion parameter model that learns to imitate the reasoning process of LFMs. Orca learns from rich signals from GPT-4 including explanation traces; step-by-step thought processes; and other complex instructions, guided by teacher assistance from ChatGPT. To promote this progressive learning, we tap into large-scale and diverse imitation data with judicious sampling and selection. Orca surpasses conventional state-of-the-art instruction-tuned models such as Vicuna-13B by more than 100% in complex zero-shot reasoning benchmarks like Big-Bench Hard (BBH) and 42% on AGIEval. Moreover, Orca reaches parity with ChatGPT on the BBH benchmark and shows competitive performance (4 pts gap with optimized system message) in professional and academic examinations like the SAT, LSAT, GRE, and GMAT, both in zero-shot settings without CoT; while trailing behind GPT-4. Our research indicates that learning from step-by-step explanations, whether these are generated by humans or more advanced AI models, is a promising direction to improve model capabilities and skills.

Generative language models are promising for assisting human writing in various domains. This manuscript aims to build generative language models in the patent domain and evaluate model performance from a human-centric perspective. The perspective is to measure the ratio of keystrokes that can be saved by autocompletion based on generative patent language models. A higher ratio means a more effective model which can save more keystrokes. This metric can be used to benchmark model performance. The metric is different from conventional machine-centric metrics that are token-based instead of keystroke-based. In terms of model size, the largest model built in this manuscript is 6B, which is state-of-the-art in the patent domain. Based on the metric, it is found that the largest model is not necessarily the best for the human-centric metric. The finding means that keeping increasing model sizes in the patent domain might be unnecessary if the purpose is to assist human writing with autocompletion. Several patent language models are pre-trained from scratch in this research. The pre-trained models are released for future researchers. Several visualization tools are also provided. The importance of building a generative language model in the patent domain is the potential to facilitate creativity and innovations in the future.

Researchers usually come up with new ideas only after thoroughly comprehending vast quantities of literature. The difficulty of this procedure is exacerbated by the fact that the number of academic publications is growing exponentially. In this study, we devise a framework based on concept co-occurrence for academic idea inspiration, which has been integrated into a research assistant system. From our perspective, the fusion of two concepts that co-occur in an academic paper can be regarded as an important way of the emergence of a new idea. We construct evolving concept graphs according to the co-occurrence relationship of concepts from 20 disciplines or topics. Then we design a temporal link prediction method based on masked language model to explore potential connections between different concepts. To verbalize the newly discovered connections, we also utilize the pretrained language model to generate a description of an idea based on a new data structure called co-occurrence citation quintuple. We evaluate our proposed system using both automatic metrics and human assessment. The results demonstrate that our system has broad prospects and can assist researchers in expediting the process of discovering new ideas.

Neural classifiers are non linear systems providing decisions on the classes of patterns, for a given problem they have learned. The output computed by a classifier for each pattern constitutes an approximation of the output of some unknown function, mapping pattern data to their respective classes. The lack of knowledge of such a function along with the complexity of neural classifiers, especially when these are deep learning architectures, do not permit to obtain information on how specific predictions have been made. Hence, these powerful learning systems are considered as black boxes and in critical applications their use tends to be considered inappropriate. Gaining insight on such a black box operation constitutes a one way approach in interpreting operation of neural classifiers and assessing the validity of their decisions. In this paper we tackle this problem introducing a novel methodology for discovering which features are considered relevant by a trained neural classifier and how they affect the classifier's output, thus obtaining an explanation on its decision. Although, feature relevance has received much attention in the machine learning literature here we reconsider it in terms of nonlinear parameter estimation targeted by a set membership approach which is based on interval analysis. Hence, the proposed methodology builds on sound mathematical approaches and the results obtained constitute a reliable estimation of the classifier's decision premises.

Numerous algorithms have been produced for the fundamental problem of clustering under many different notions of fairness. Perhaps the most common family of notions currently studied is group fairness, in which proportional group representation is ensured in every cluster. We extend this direction by considering the downstream application of clustering and how group fairness should be ensured for such a setting. Specifically, we consider a common setting in which a decision-maker runs a clustering algorithm, inspects the center of each cluster, and decides an appropriate outcome (label) for its corresponding cluster. In hiring for example, there could be two outcomes, positive (hire) or negative (reject), and each cluster would be assigned one of these two outcomes. To ensure group fairness in such a setting, we would desire proportional group representation in every label but not necessarily in every cluster as is done in group fair clustering. We provide algorithms for such problems and show that in contrast to their NP-hard counterparts in group fair clustering, they permit efficient solutions. We also consider a well-motivated alternative setting where the decision-maker is free to assign labels to the clusters regardless of the centers' positions in the metric space. We show that this setting exhibits interesting transitions from computationally hard to easy according to additional constraints on the problem. Moreover, when the constraint parameters take on natural values we show a randomized algorithm for this setting that always achieves an optimal clustering and satisfies the fairness constraints in expectation. Finally, we run experiments on real world datasets that validate the effectiveness of our algorithms.

This paper is a submission to the Open Philanthropy AI Worldviews Contest. In it, we estimate the likelihood of transformative artificial general intelligence (AGI) by 2043 and find it to be <1%. Specifically, we argue: The bar is high: AGI as defined by the contest - something like AI that can perform nearly all valuable tasks at human cost or less - which we will call transformative AGI is a much higher bar than merely massive progress in AI, or even the unambiguous attainment of expensive superhuman AGI or cheap but uneven AGI. Many steps are needed: The probability of transformative AGI by 2043 can be decomposed as the joint probability of a number of necessary steps, which we group into categories of software, hardware, and sociopolitical factors. No step is guaranteed: For each step, we estimate a probability of success by 2043, conditional on prior steps being achieved. Many steps are quite constrained by the short timeline, and our estimates range from 16% to 95%. Therefore, the odds are low: Multiplying the cascading conditional probabilities together, we estimate that transformative AGI by 2043 is 0.4% likely. Reaching >10% seems to require probabilities that feel unreasonably high, and even 3% seems unlikely. Thoughtfully applying the cascading conditional probability approach to this question yields lower probability values than is often supposed. This framework helps enumerate the many future scenarios where humanity makes partial but incomplete progress toward transformative AGI.

This research delves into the current literature on bias in Natural Language Processing Models and the techniques proposed to mitigate the problem of bias, including why it is important to tackle bias in the first place. Additionally, these techniques are further analysed in the light of newly developed models that tower in size over past editions. To achieve those aims, the authors of this paper conducted their research on GPT3 by OpenAI, the largest NLP model available to consumers today. With 175 billion parameters in contrast to BERTs 340 million, GPT3 is the perfect model to test the common pitfalls of NLP models. Tests were conducted through the development of an Applicant Tracking System using GPT3. For the sake of feasibility and time constraints, the tests primarily focused on gender bias, rather than all or multiple types of bias. Finally, current mitigation techniques are considered and tested to measure their degree of functionality.

Generating a Chain of Thought (CoT) has been shown to consistently improve large language model (LLM) performance on a wide range of NLP tasks. However, prior work has mainly focused on logical reasoning tasks (e.g. arithmetic, commonsense QA); it remains unclear whether improvements hold for more diverse types of reasoning, especially in socially situated contexts. Concretely, we perform a controlled evaluation of zero-shot CoT across two socially sensitive domains: harmful questions and stereotype benchmarks. We find that zero-shot CoT reasoning in sensitive domains significantly increases a model's likelihood to produce harmful or undesirable output, with trends holding across different prompt formats and model variants. Furthermore, we show that harmful CoTs increase with model size, but decrease with improved instruction following. Our work suggests that zero-shot CoT should be used with caution on socially important tasks, especially when marginalized groups or sensitive topics are involved.

Federated learning (FL) is a trending distributed learning framework that enables collaborative model training without data sharing. Machine learning models trained on datasets can potentially expose the private information of the training data, revealing details about individual data records. In this study, we focus on the FL paradigm that grants clients the ``right to be forgotten''. The forgettable FL framework should bleach its global model weights as it has never seen that client and hence does not reveal any information about the client. To this end, we propose the Forgettable Federated Linear Learning (2F2L) framework featured with novel training and data removal strategies. The training pipeline, named Federated linear training, employs linear approximation on the model parameter space to enable our 2F2L framework work for deep neural networks while achieving comparable results with canonical neural network training. We also introduce FedRemoval, an efficient and effective removal strategy that tackles the computational challenges in FL by approximating the Hessian matrix using public server data from the pretrained model. Unlike the previous uncertified and heuristic machine unlearning methods in FL, we provide theoretical guarantees by bounding the differences of model weights by our FedRemoval and that from retraining from scratch. Experimental results on MNIST and Fashion-MNIST datasets demonstrate the effectiveness of our method in achieving a balance between model accuracy and information removal, outperforming baseline strategies and approaching retraining from scratch.