Pre-training Large Language Models (LLMs) require massive amounts of text data, and the performance of the LLMs typically correlates with the scale and quality of the datasets. This means that it may be challenging to build LLMs for smaller languages such as Nordic ones, where the availability of text corpora is limited. In order to facilitate the development of the LLMS in the Nordic languages, we curate a high-quality dataset consisting of 1.2TB of text, in all of the major North Germanic languages (Danish, Icelandic, Norwegian, and Swedish), as well as some high-quality English data. This paper details our considerations and processes for collecting, cleaning, and filtering the dataset.

In this work, we induce character-level noise in various forms when fine-tuning BERT to enable zero-shot cross-lingual transfer to unseen dialects and languages. We fine-tune BERT on three sentence-level classification tasks and evaluate our approach on an assortment of unseen dialects and languages. We find that character-level noise can be an extremely effective agent of cross-lingual transfer under certain conditions, while it is not as helpful in others. Specifically, we explore these differences in terms of the nature of the task and the relationships between source and target languages, finding that introduction of character-level noise during fine-tuning is particularly helpful when a task draws on surface level cues and the source-target cross-lingual pair has a relatively high lexical overlap with shorter (i.e., less meaningful) unseen tokens on average.

Large pre-trained code generation models, such as OpenAI Codex, can generate syntax- and function-correct code, making the coding of programmers more productive and our pursuit of artificial general intelligence closer. In this paper, we introduce CodeGeeX, a multilingual model with 13 billion parameters for code generation. CodeGeeX is pre-trained on 850 billion tokens of 23 programming languages as of June 2022. Our extensive experiments suggest that CodeGeeX outperforms multilingual code models of similar scale for both the tasks of code generation and translation on HumanEval-X. Building upon HumanEval (Python only), we develop the HumanEval-X benchmark for evaluating multilingual models by hand-writing the solutions in C++, Java, JavaScript, and Go. In addition, we build CodeGeeX-based extensions on Visual Studio Code, JetBrains, and Cloud Studio, generating 4.7 billion tokens for tens of thousands of active users per week. Our user study demonstrates that CodeGeeX can help to increase coding efficiency for 83.4% of its users. Finally, CodeGeeX is publicly accessible and in Sep. 2022, we open-sourced its code, model weights (the version of 850B tokens), API, extensions, and HumanEval-X at https://github.com/THUDM/CodeGeeX.

The popularity of generative text AI tools in answering questions has led to concerns regarding their potential negative impact on students' academic performance and the challenges that educators face in evaluating student learning. To address these concerns, this paper introduces an evolutionary approach that aims to identify the best set of Bloom's taxonomy keywords to generate questions that these tools have low confidence in answering. The effectiveness of this approach is evaluated through a case study that uses questions from a Data Structures and Representation course being taught at the University of New South Wales in Canberra, Australia. The results demonstrate that the optimization algorithm is able to find keywords from different cognitive levels to create questions that ChatGPT has low confidence in answering. This study is a step forward to offer valuable insights for educators seeking to create more effective questions that promote critical thinking among students.

Several decision problems that are encountered in various business domains can be modeled as mathematical programs, i.e. optimization problems. The process of conducting such modeling often requires the involvement of experts trained in operations research and advanced algorithms. Surprisingly, despite the significant advances in the methods for program and code synthesis, AutoML, learning to optimize etc., there has been little or no attention paid to automating the task of synthesizing mathematical programs. We imagine a scenario where the specifications for modeling, i.e. the objective and constraints are expressed in an unstructured form in natural language (NL) and the mathematical program has to be synthesized from such an NL specification. In this work we evaluate the efficacy of employing CodeT5 with data augmentation and post-processing of beams. We utilize GPT-3 with back translation for generation of synthetic examples. Further we apply rules of linear programming to score beams and correct beams based on common error patterns. We observe that with these enhancements CodeT5 base gives an execution accuracy of 0.73 which is significantly better than zero-shot execution accuracy of 0.41 by ChatGPT and 0.36 by Codex.

Decoding human visual neural representations is a challenging task with great scientific significance in revealing vision-processing mechanisms and developing brain-like intelligent machines. Most existing methods are difficult to generalize to novel categories that have no corresponding neural data for training. The two main reasons are 1) the under-exploitation of the multimodal semantic knowledge underlying the neural data and 2) the small number of paired (stimuli-responses) training data. To overcome these limitations, this paper presents a generic neural decoding method called BraVL that uses multimodal learning of brain-visual-linguistic features. We focus on modeling the relationships between brain, visual and linguistic features via multimodal deep generative models. Specifically, we leverage the mixture-of-product-of-experts formulation to infer a latent code that enables a coherent joint generation of all three modalities. To learn a more consistent joint representation and improve the data efficiency in the case of limited brain activity data, we exploit both intra- and inter-modality mutual information maximization regularization terms. In particular, our BraVL model can be trained under various semi-supervised scenarios to incorporate the visual and textual features obtained from the extra categories. Finally, we construct three trimodal matching datasets, and the extensive experiments lead to some interesting conclusions and cognitive insights: 1) decoding novel visual categories from human brain activity is practically possible with good accuracy; 2) decoding models using the combination of visual and linguistic features perform much better than those using either of them alone; 3) visual perception may be accompanied by linguistic influences to represent the semantics of visual stimuli. Code and data: https://github.com/ChangdeDu/BraVL.

This study focuses on using large language models (LLMs) as a planner for embodied agents that can follow natural language instructions to complete complex tasks in a visually-perceived environment. The high data cost and poor sample efficiency of existing methods hinders the development of versatile agents that are capable of many tasks and can learn new tasks quickly. In this work, we propose a novel method, LLM-Planner, that harnesses the power of large language models to do few-shot planning for embodied agents. We further propose a simple but effective way to enhance LLMs with physical grounding to generate and update plans that are grounded in the current environment. Experiments on the ALFRED dataset show that our method can achieve very competitive few-shot performance: Despite using less than 0.5% of paired training data, LLM-Planner achieves competitive performance with recent baselines that are trained using the full training data. Existing methods can barely complete any task successfully under the same few-shot setting. Our work opens the door for developing versatile and sample-efficient embodied agents that can quickly learn many tasks. Website: https://dki-lab.github.io/LLM-Planner

The training of modern large language models (LLMs) takes place in a regime where most training examples are seen only a few times by the model during the course of training. What does a model remember about such examples seen only a few times during training and how long does that memory persist in the face of continuous training with new examples? Here, we investigate these questions through simple recognition, recall, and retention experiments with LLMs. In recognition experiments, we ask if the model can distinguish the seen example from a novel example; in recall experiments, we ask if the model can correctly recall the seen example when cued by a part of it; and in retention experiments, we periodically probe the model's memory for the original examples as the model is trained continuously with new examples. We find that a single exposure is generally sufficient for a model to achieve near perfect accuracy even in very challenging recognition experiments. We estimate that the recognition performance of even small language models easily exceeds human recognition performance reported in similar experiments with humans (Shepard, 1967). Achieving near perfect recall takes more exposures, but most models can do it in just 3 exposures. The flip side of this remarkable capacity for fast learning is that precise memories are quickly overwritten: recall performance for the original examples drops steeply over the first 10 training updates with new examples, followed by a more gradual decline. Even after 100K updates, however, some of the original examples are still recalled near perfectly. A qualitatively similar retention pattern has been observed in human long-term memory retention studies before (Bahrick, 1984). Finally, recognition is much more robust to interference than recall and memory for natural language sentences is generally superior to memory for stimuli without structure.

Apple cofounder Steve Wozniak has joined Elon Musk and leading AI academics in calling for a pause in advanced AI development. The letter says that current AI development is out of control, and may pose "profound risks to society and humanity." As stated in the widely-endorsed Asilomar AI Principles, Advanced AI could represent a profound change in the history of life on Earth, and should be planned for and managed with commensurate care and resources. Unfortunately, this level of planning and management is not happening, even though recent months have seen AI labs locked in an out-of-control race to develop and deploy ever more powerful digital minds that no one – not even their creators – can understand, predict, or reliably control. It says that AI systems put a great many jobs at risk, and asks a number of questions.

"How well does the AI perform clinically? And my answer is, I'm stunned to say: Better than many doctors I've observed."--Isaac The large language model GPT-4 (LLM, aka generative AI chatbot or foundation model) was just released 2 weeks ago (14 March) but there's already been much written about its advance beyond ChatGPT, released 30 November 2022, "the most successful new product in the history of the western world" with over 100 million users in just 2 months. A new book by Peter Lee, Carey Goldberg, Isaac Kohane will be released as an e-book April 15th and as a paperback May 3rd and I've had the chance to read it. With my keen interest for how AI can transform medicine (as written about in Deep Medicine and multiple recent review papers here, here, here), I couldn't put it down.