Large language models (LLMs) often hallucinate and lack the ability to provide attribution for their generations. Semi-parametric LMs, such as kNN-LM, approach these limitations by refining the output of an LM for a given prompt using its nearest neighbor matches in a non-parametric data store. However, these models often exhibit slow inference speeds and produce non-fluent texts. In this paper, we introduce Nearest Neighbor Speculative Decoding (NEST), a novel semi-parametric language modeling approach that is capable of incorporating real-world text spans of arbitrary length into the LM generations and providing attribution to their sources. NEST performs token-level retrieval at each inference step to compute a semi-parametric mixture distribution and identify promising span continuations in a corpus. It then uses an approximate speculative decoding procedure that accepts a prefix of the retrieved span or generates a new token. NEST significantly enhances the generation quality and attribution rate of the base LM across a variety of knowledge-intensive tasks, surpassing the conventional kNN-LM method and performing competitively with in-context retrieval augmentation. In addition, NEST substantially improves the generation speed, achieving a 1.8x speedup in inference time when applied to Llama-2-Chat 70B.

Large language models (LLMs) have shown great promise at generating robot programs from natural language given domain-specific robot application programming interfaces (APIs). However, the performance gap between proprietary LLMs and smaller open-weight LLMs remains wide. This raises a question: Can we fine-tune smaller open-weight LLMs for generating domain-specific robot programs to close the performance gap with proprietary LLMs? While Self-Instruct is a promising solution by generating a diverse set of training data, it cannot verify the correctness of these programs. In contrast, a robot simulator with a well-defined world can identify execution errors but limits the diversity of programs that it can verify. In this work, we introduce Robo-Instruct, which brings the best of both worlds -- it promotes the diversity of Self-Instruct while providing the correctness of simulator-based checking. Robo-Instruct introduces RoboSim to synthesize a consistent world state on the fly by inferring properties relevant to the program being checked, and simulating actions accordingly. Furthermore, the instructions and programs generated by Self-Instruct may be subtly inconsistent -- such as the program missing a step implied by the instruction. Robo-Instruct further addresses this with InstAlign, an instruction-program alignment procedure that revises the task instruction to reflect the actual results of the generated program. Given a few seed task descriptions and the robot APIs, Robo-Instruct is capable of generating a training dataset using only a small open-weight model. This dataset can then be used to fine-tune small open-weight language models, enabling them to match or even exceed the performance of several proprietary LLMs, such as GPT-3.5-Turbo and Gemini-Pro.

Counting is a fundamental example of generalization, whether viewed through the mathematical lens of Peano's axioms defining the natural numbers or the cognitive science literature for children learning to count. The argument holds for both cases that learning to count means learning to count infinitely. While few papers have tried to distill transformer "reasoning" to the simplest case of counting, investigating length generalization does occur throughout the literature. In the "train short, test long" paradigm of NLP, length refers to the training sentence length. In formal language recognition, length refers to the input sequence length, or the maximum stack size induced by a pushdown automata. In general problem solving, length refers to the number of hops in a deductive reasoning chain or the recursion depth. For all cases, counting is central to task success. And crucially, generalizing counting inductively is central to success on OOD instances. This work provides extensive empirical results on training language models to count. We experiment with architectures ranging from RNNs, Transformers, State-Space Models and RWKV. We present carefully-designed task formats, auxiliary tasks and positional embeddings to avoid limitations in generalization with OOD-position and OOD-vocabulary. We find that while traditional RNNs trivially achieve inductive counting, Transformers have to rely on positional embeddings to count out-of-domain. As counting is the basis for many arguments concerning the expressivity of Transformers, our finding calls for the community to reexamine the application scope of primitive functions defined in formal characterizations. Finally, modern RNNs also largely underperform traditional RNNs in generalizing counting inductively. We discuss how design choices that enable parallelized training of modern RNNs cause them to lose merits of a recurrent nature.

We present ESG-FTSE, the first corpus comprised of news articles with Environmental, Social and Governance (ESG) relevance annotations. In recent years, investors and regulators have pushed ESG investing to the mainstream due to the urgency of climate change. This has led to the rise of ESG scores to evaluate an investment's credentials as socially responsible. While demand for ESG scores is high, their quality varies wildly. Quantitative techniques can be applied to improve ESG scores, thus, responsible investing. To contribute to resource building for ESG and financial text mining, we pioneer the ESG-FTSE corpus. We further present the first of its kind ESG annotation schema. It has three levels: a binary classification (relevant versus irrelevant news articles), ESG classification (ESG-related news articles), and target company. Both supervised and unsupervised learning experiments for ESG relevance detection were conducted to demonstrate that the corpus can be used in different settings to derive accurate ESG predictions.

Proteins are essential for almost all biological processes and derive their diverse functions from complex 3D structures, which are in turn determined by their amino acid sequences. In this paper, we exploit the rich biological inductive bias of amino acid sequences and introduce FoldFlow-2, a novel sequence-conditioned SE(3)-equivariant flow matching model for protein structure generation. FoldFlow-2 presents substantial new architectural features over the previous FoldFlow family of models including a protein large language model to encode sequence, a new multi-modal fusion trunk that combines structure and sequence representations, and a geometric transformer based decoder. To increase diversity and novelty of generated samples -- crucial for de-novo drug design -- we train FoldFlow-2 at scale on a new dataset that is an order of magnitude larger than PDB datasets of prior works, containing both known proteins in PDB and high-quality synthetic structures achieved through filtering. We further demonstrate the ability to align FoldFlow-2 to arbitrary rewards, e.g. increasing secondary structures diversity, by introducing a Reinforced Finetuning (ReFT) objective. We empirically observe that FoldFlow-2 outperforms previous state-of-the-art protein structure-based generative models, improving over RFDiffusion in terms of unconditional generation across all metrics including designability, diversity, and novelty across all protein lengths, as well as exhibiting generalization on the task of equilibrium conformation sampling. Finally, we demonstrate that a fine-tuned FoldFlow-2 makes progress on challenging conditional design tasks such as designing scaffolds for the VHH nanobody.

In spoken dialogue, even if two current turns are the same sentence, their responses might still differ when they are spoken in different styles. The spoken styles, containing paralinguistic and prosodic information, mark the most significant difference between text and speech modality. When using text-only LLMs to model spoken dialogue, text-only LLMs cannot give different responses based on the speaking style of the current turn. In this paper, we focus on enabling LLMs to listen to the speaking styles and respond properly. Our goal is to teach the LLM that "even if the sentences are identical if they are spoken in different styles, their corresponding responses might be different". Since there is no suitable dataset for achieving this goal, we collect a speech-to-speech dataset, StyleTalk, with the following desired characteristics: when two current speeches have the same content but are spoken in different styles, their responses will be different. To teach LLMs to understand and respond properly to the speaking styles, we propose the Spoken-LLM framework that can model the linguistic content and the speaking styles. We train Spoken-LLM using the StyleTalk dataset and devise a two-stage training pipeline to help the Spoken-LLM better learn the speaking styles. Based on extensive experiments, we show that Spoken-LLM outperforms text-only baselines and prior speech LLMs methods.

Recent years have seen the dramatic rise of the usage of AI algorithms in pure mathematics and fundamental sciences such as theoretical physics. This is perhaps counter-intuitive since mathematical sciences require the rigorous definitions, derivations, and proofs, in contrast to the experimental sciences which rely on the modelling of data with error-bars. In this Perspective, we categorize the approaches to mathematical discovery as "top-down", "bottom-up" and "meta-mathematics", as inspired by historical examples. We review some of the progress over the last few years, comparing and contrasting both the advances and the short-comings in each approach. We argue that while the theorist is in no way in danger of being replaced by AI in the near future, the hybrid of human expertise and AI algorithms will become an integral part of theoretical discovery.

This paper analyzes Large Language Models (LLMs) with regard to their programming exercise generation capabilities. Through a survey study, we defined the state of the art, extracted their strengths and weaknesses and finally proposed an evaluation matrix, helping researchers and educators to decide which LLM is the best fitting for the programming exercise generation use case. We also found that multiple LLMs are capable of producing useful programming exercises. Nevertheless, there exist challenges like the ease with which LLMs might solve exercises generated by LLMs. This paper contributes to the ongoing discourse on the integration of LLMs in education.

Yoshimura origami is a classical folding pattern that has inspired many deployable structure designs. Its applications span from space exploration, kinetic architectures, and soft robots to even everyday household items. However, despite its wide usage, Yoshimura has been fixated on a set of design constraints to ensure its flat-foldability. Through extensive kinematic analysis and prototype tests, this study presents a new Yoshimura that intentionally defies these constraints. Remarkably, one can impart a unique meta-stability by using the Golden Ratio angle to define the triangular facets of a generalized Yoshimura. As a result, when its facets are strategically popped out, a ``Golden Ratio Yoshimura'' boom with $m$ modules can be theoretically reconfigured into $8^m$ geometrically unique and load-bearing shapes. This result not only challenges the existing design norms but also opens up a new avenue to create deployable and versatile structural systems.

Most previous studies conduct factor analysis within the framework of incident-related label prediction, where the developed models can be categorized into short-term and long-term prediction models. The long-term models offer a more strategic approach, enabling more proactive risk management, compared to the short-term ones. Nevertheless, few studies have devoted to rigorously identifying the key factors for the long-term prediction and undertaking comprehensive factor analysis. Hence, this study aims to delve into the key factors for predicting the incident risk levels in the subsequent year given a specific datestamp. The majority of candidate factors potentially contributing to the incident risk are collected from vessels' historical safety performance data spanning up to five years. An improved embedded feature selection method, which integrates Random Forest classifier with a feature filtering process, is proposed to identify key risk-contributing factors from the candidate pool. A dataset with information of 131,652 vessels collected from 2015 to 2023 is utilized for case study. The results demonstrate superior performances of the proposed method in incident prediction and factor interpretability. Comprehensive analysis is conducted upon the key factors, which could help maritime stakeholders formulate management strategies for incident prevention.