A long-standing ambition for artificial intelligence (AI) in biomedicine is the development of AI systems that could eventually make major scientific discoveries, with the potential to be worthy of a Nobel Prize--fulfilling the Nobel Turing Challenge [1]. While the concept of an "AI scientist" is aspirational, advances in agent-based AI pave the way to the development of AI agents as conversable systems capable of skeptical learning and reasoning that coordinate large language models (LLMs), machine learning (ML) tools, experimental platforms, or even combinations of them [2-5] (Figure 1). The complexity of biological problems requires a multistage approach, where decomposing complex questions into simpler tasks is necessary. AI agents can break down a problem into manageable subtasks, which can then be addressed by agents with specialized functions for targeted problem-solving and integration of scientific knowledge, paving the way toward a future in which a major biomedical discovery is made solely by AI [2, 6].

Large Language Models (LLMs) have demonstrated impressive abilities in recent years with regards to code generation and understanding. However, little work has investigated how documentation and other code properties affect an LLM's ability to understand and generate code or documentation. We present an empirical analysis of how underlying properties of code or documentation can affect an LLM's capabilities. We show that providing an LLM with "incorrect" documentation can greatly hinder code understanding, while incomplete or missing documentation does not seem to significantly affect an LLM's ability to understand code.

With the primary focus on evaluating the effectiveness of large language models for automatic reference-less translation assessment, this work presents our experiments on mimicking human direct assessment to evaluate the quality of translations in English and Indian languages. We constructed a translation evaluation task where we performed zero-shot learning, in-context example-driven learning, and fine-tuning of large language models to provide a score out of 100, where 100 represents Figure 1: Spearman co-relation: Human translation a perfect translation and 1 represents a poor evaluation vs different reference-less translation translation. We compared the performance of evaluation metrics. Llama-2-7b-Adapt (lora), our trained systems with existing methods such Llama-2-13b-Adapt (lora), Mistral-7b-Adpt (lora), as COMET, BERT-Scorer, and LABSE, and COMET-QE (https://github.com/Unbabel/COMET)

For many years, different fields in applied machine learning such as computer vision, speech or natural language processing have been building domainspecialised solutions, mainly fueled by the need for hand-crafted features to then be used with classical machine learning methods. Currently, we are witnessing an opposing trend towards developing more generalist architectures, driven by Large Language Models and multi-modal foundational models. These architectures are designed to tackle a variety of tasks, including those that were previously unseen and using inputs across multiple modalities. Taking this trend of generalization to the extreme suggests the possibility of a single deep network architecture capable of solving all tasks, rather than just one or a limited subset as currently observed. This position paper aims to explore the idea that developing such a unified architecture may not be as difficult as one may think and propose a theoretical framework of how it could be constructed. Our proposal is based on the following assumptions. Firstly, the solution to any given task can be expressed through a sequence of instructions, as most tasks necessitate implementation in code on conventional computing hardware, which operates inherently in a sequential manner. Second, recent Generative AI, especially Transformer-based models, demonstrate not only a leap in performance but also a potential as an architecture capable of constructing algorithms for a wide range of domains. For example, GPT-4 shows exceptional capability at in-context learning of novel tasks which is hard to explain in any other way than the ability to compose novel solutions from fragments on previously learnt algorithms. Third, the observation that the main missing component in developing a truly generalised network is an efficient approach for self-consistent input of previously learnt sub-steps of an algorithm and their (implicit) composition during the network's internal forward pass.

The computational intensity of Large Language Models (LLMs) is a critical bottleneck, primarily due to the $O(n^2)$ complexity of the attention mechanism in transformer architectures. Addressing this, sparse attention emerges as a key innovation, aiming to reduce computational load while maintaining model performance. This study presents a rigorous theoretical analysis of the sparsity in attention scores within LLMs, particularly under the framework of Gaussian inputs. By establishing a set of foundational assumptions and employing a methodical theoretical approach, we unravel the intrinsic characteristics of attention score sparsity and its implications on computational efficiency. Our main contribution lies in providing a detailed theoretical examination of how sparsity manifests in attention mechanisms, offering insights into the potential trade-offs between computational savings and model effectiveness. This work not only advances our understanding of sparse attention but also provides a scaffold for future research in optimizing the computational frameworks of LLMs, paving the way for more scalable and efficient AI systems.

The pervasive influence of social biases in language data has sparked the need for benchmark datasets that capture and evaluate these biases in Large Language Models (LLMs). Existing efforts predominantly focus on English language and the Western context, leaving a void for a reliable dataset that encapsulates India's unique socio-cultural nuances. To bridge this gap, we introduce IndiBias, a comprehensive benchmarking dataset designed specifically for evaluating social biases in the Indian context. We filter and translate the existing CrowS-Pairs dataset to create a benchmark dataset suited to the Indian context in Hindi language. Additionally, we leverage LLMs including ChatGPT and InstructGPT to augment our dataset with diverse societal biases and stereotypes prevalent in India. The included bias dimensions encompass gender, religion, caste, age, region, physical appearance, and occupation. We also build a resource to address intersectional biases along three intersectional dimensions. Our dataset contains 800 sentence pairs and 300 tuples for bias measurement across different demographics. The dataset is available in English and Hindi, providing a size comparable to existing benchmark datasets. Furthermore, using IndiBias we compare ten different language models on multiple bias measurement metrics. We observed that the language models exhibit more bias across a majority of the intersectional groups.

Large language models (LLMs) often generate content that contains factual errors when responding to fact-seeking prompts on open-ended topics. To benchmark a model's long-form factuality in open domains, we first use GPT-4 to generate LongFact, a prompt set comprising thousands of questions spanning 38 topics. We then propose that LLM agents can be used as automated evaluators for long-form factuality through a method which we call Search-Augmented Factuality Evaluator (SAFE). SAFE utilizes an LLM to break down a long-form response into a set of individual facts and to evaluate the accuracy of each fact using a multi-step reasoning process comprising sending search queries to Google Search and determining whether a fact is supported by the search results. Furthermore, we propose extending F1 score as an aggregated metric for long-form factuality. To do so, we balance the percentage of supported facts in a response (precision) with the percentage of provided facts relative to a hyperparameter representing a user's preferred response length (recall). Empirically, we demonstrate that LLM agents can outperform crowdsourced human annotators - on a set of ~16k individual facts, SAFE agrees with crowdsourced human annotators 72% of the time, and on a random subset of 100 disagreement cases, SAFE wins 76% of the time. At the same time, SAFE is more than 20 times cheaper than human annotators. We also benchmark thirteen language models on LongFact across four model families (Gemini, GPT, Claude, and PaLM-2), finding that larger language models generally achieve better long-form factuality. LongFact, SAFE, and all experimental code are available at https://github.com/google-deepmind/long-form-factuality.

Mixture-of-Expert (MoE) based large language models (LLMs), such as the recent Mixtral and DeepSeek-MoE, have shown great promise in scaling model size without suffering from the quadratic growth of training cost of dense transformers. Like dense models, training MoEs requires answering the same question: given a training budget, what is the optimal allocation on the model size and number of tokens? We study the scaling law of MoE-based LLMs regarding the relations between the model performance, model size, dataset size, and the expert degree. Echoing previous research studying MoE in different contexts, we observe the diminishing return of increasing the number of experts, but this seems to suggest we should scale the number of experts until saturation, as the training cost would remain constant, which is problematic during inference time. We propose to amend the scaling law of MoE by introducing inference efficiency as another metric besides the validation loss. We find that MoEs with a few (4/8) experts are the most serving efficient solution under the same performance, but costs 2.5-3.5x more in training. On the other hand, training a (16/32) expert MoE much smaller (70-85%) than the loss-optimal solution, but with a larger training dataset is a promising setup under a training budget.

AI-based systems have the potential to assist employers with many aspects of human resources (HR) management, from benefits administration to coaching and development to its most common HR use case, applicant screening. The global HR technology market based on predictive models was already rapidly growing prior to 2022, but attention to AI tools received a dramatic boost with the advent of large language models (LLMs), which are models that are highly adept at understanding, summarizing, and evaluating text data. Given the primacy of text data in the job application process, an emerging HR use case for modern LLMs is to ingest entire application dossiers--including resumes, essays, and transcripts captured from interviews--and output seemingly cogent assessments of candidates' qualifications. As hiring use cases proliferate, however, employers and policymakers are racing to establish guidelines around whether the algorithmic evaluation of candidates comports with employment discrimination law, and how to audit commonly deployed AI tools to ensure they are not discriminatory. The ethical and legal implications of using predictive tools in HR has motivated a body of academic work (Raghavan et al., 2020; Tambe et al., 2019). Policymakers have matched the attention of firms and researchers, introducing a wave of legislation governing high-stakes algorithmic decision making, and hiring in particular (e.g., New York LL 144 or Illinois 820 ILCS 42).

Large language models (LLMs) have achieved significant performance in various natural language reasoning tasks. However, they still struggle with performing first-order logic reasoning over formal logical theories expressed in natural language. This is because the previous LLMs-based reasoning systems have the theoretical incompleteness issue. As a result, it can only address a limited set of simple reasoning problems, which significantly decreases their generalization ability. To address this issue, we propose a novel framework, named Generalizable and Faithful Reasoner (GFaiR), which introduces the paradigm of resolution refutation. Resolution refutation has the capability to solve all first-order logic reasoning problems by extending reasoning rules and employing the principle of proof by contradiction, so our system's completeness can be improved by introducing resolution refutation. Experimental results demonstrate that our system outperforms previous works by achieving state-of-the-art performances in complex scenarios while maintaining performances in simple scenarios. Besides, we observe that GFaiR is faithful to its reasoning process.