Reviews & Analysis

Users often overestimate the accuracy of large language models (LLMs). A new approach examines user perceptions and finds that aligning LLM explanations with the models’ internal confidence improves user perception.

The immunogenic binding interactions of antigens are complex and interconnected. A new transformer-based model can simultaneously predict the bindings of antigens to two main receptors.

The development of comprehensive benchmarks to assess the performance of algorithms on causal tasks is an important, emerging area. The introduction of two physical ‘causal chamber’ systems serves as a firm step towards future, more reliable benchmarks in the field.

Machine unlearning techniques remove undesirable data and associated model capabilities while preserving essential knowledge, so that machine learning models can be updated without costly retraining. Liu et al. review recent advances and opportunities in machine unlearning in LLMs, revisiting methodologies and overlooked principles for future improvements and exploring emerging applications in copyright and privacy safeguards and in reducing sociotechnical harms.

The performance of omics prediction models can be significantly improved by combining limited patient proteomic data with widely available electronic health records.

With widespread generation and availability of synthetic data, AI systems are increasingly trained on their own outputs, leading to various technical and ethical challenges. The authors analyse this development and discuss measures to mitigate the potential adverse effects of ‘AI eating itself’.

Large general-purpose models are becoming more prevalent and useful, but also harder to train and find suitable training data for. Zheng et al. discuss how models can be used to train other models.

As powerful institutions increasingly promote AI systems, efforts to align those systems with human morality have grown. An open-source AI system aims to predict human moral judgments across a broad spectrum of everyday situations expressed in natural language. Identifying the limitations of such systems offers important insights for future work.

Tackling partial differential equations with machine learning solvers is a promising direction, but recent analysis reveals challenges with making fair comparisons to previous methods. Stronger benchmark problems are needed for the field to advance.

Social learning is a powerful strategy of adaptation in nature. An interactive rat-like robot that engages in imitation learning with a freely behaving rat opens a way to study social behaviours.

A deep learning-based method shows promise in issuing early warnings of rate-induced tipping, of particular interest in anticipating effects due to anthropogenic climate change.

Recurrent neural networks are efficient and capable agents for discovering new peptides with strong self-organizing capabilities.

A self-decoupling tactile sensor dramatically reduces calibration time for three-dimensional force measurement, scaling from cubic (N³) to linear (3N). This advancement facilitates robotic tactile perception in human–machine interfaces.

The wide adoption of AI in biomedical research raises concerns about misuse risks. Trotsyuk, Waeiss et al. propose a framework that provides a starting point for researchers to consider how risks specific to their work could be mitigated, using existing ethical frameworks, regulatory measures and off-the-shelf AI solutions.

Schmidgall et al. describe a pathway for building general-purpose machine learning models for robot-assisted surgery, including mechanisms for avoiding risk and handing over control to surgeons, and improving safety and outcomes beyond demonstration data.

Forecasting epidemic progression is a complex task influenced by various factors, including human behaviour, pathogen dynamics and environmental conditions. Rodríguez, Kamarthi and colleagues provide a review of machine learning methods for epidemic forecasting from a data-centric computational perspective.

Using deep reinforcement learning, flexible skills and behaviours emerge in humanoid robots, as demonstrated in two recent reports.

Training data are crucial for advancements in artificial intelligence, but many questions remain regarding the provenance of training datasets, license enforcement and creator consent. Mahari et al. provide a set of tools for tracing, documenting and sharing AI training data and highlight the importance for developers to engage with metadata of datasets.

Large language models (LLMs) present challenges, including a tendency to produce false or misleading content and the potential to create misinformation or disinformation. Augenstein and colleagues explore issues related to factuality in LLMs and their impact on fact-checking.

Constructing spatial maps from sensory inputs is challenging in both neuroscience and artificial intelligence. A recent study demonstrates that a self-attention neural network using predictive coding can generate an environmental map in its latent space as an agent that navigates the environment.