Language and vision-language models have shown impressive performance across a wide range of tasks, but their internal mechanisms remain only partly understood. In this work, we study how individual attention heads in text-generative models specialize in specific semantic or visual attributes. Building on an established interpretability method, we reinterpret the practice of probing intermediate activations with the final decoding layer through the lens of signal processing. This lets us analyze multiple samples in a principled way and rank attention heads based on their relevance to target concepts. Our results show consistent patterns of specialization at the head level across both unimodal and multimodal transformers. Remarkably, we find that editing as few as 1% of the heads, selected using our method, can reliably suppress or enhance targeted concepts in the model output.

The potential for artificial intelligence to affect election results is a major public concern. Two new papers - with experiments conducted in four countries - demonstrate that chatbots powered by large language models (LLMs) are quite effective at political persuasion, moving opposition voters' preferences by 10 percentage points or more in many cases. The LLMs' persuasiveness comes not from being masters of psychological manipulation, but because they come up with so many claims supporting their arguments for candidates' policy positions. "LLMs can really move people's attitudes towards presidential candidates and policies, and they do it by providing many factual claims that support their side," said David Rand, a senior author on both papers. "But those claims aren't necessarily accurate - and even arguments built on accurate claims can still mislead by omission."

In the traditional supervised learning framework, active learning enables the learner to actively engage in the construction of the labeled training set by selecting a fixed-sized subset of unlabeled examples for labeling by an oracle, where the number of labels requested is referred to as the budget .

AI can influence voters' minds. What does that mean for democracy? AI chatbots may have the power to influence voters' opinions Does the persuasive power of AI chatbots spell the beginning of the end for democracy? In one of the largest surveys to date exploring how these tools can influence voter attitudes, AI chatbots were more persuasive than traditional political campaign tools including advertisements and pamphlets, and as persuasive as seasoned political campaigners. But at least some researchers identify reasons for optimism in the way in which the AI tools shifted opinions.