Trump-Xi meeting: Who has the upper hand? Could Trump go for a third term? Is the US eyeing its next Latin American target? Why is Trump tearing down parts of the White House? United States President Donald Trump and his Chinese counterpart Xi Jinping have agreed to a trade truce under which the US will ease tariffs and Beijing will restart imports of US soya beans, delay the introduction of export restrictions on some of its rare earth metals and intensify efforts to curb illegal fentanyl trafficking.

Standard training metrics like loss fail to explain the emergence of complex capabilities in large language models. We take a spectral approach to investigate the geometry of learned representations across pretraining and post-training, measuring effective rank (RankMe) and eigenspectrum decay ($α$-ReQ). With OLMo (1B-7B) and Pythia (160M-12B) models, we uncover a consistent non-monotonic sequence of three geometric phases during autoregressive pretraining. The initial "warmup" phase exhibits rapid representational collapse. This is followed by an "entropy-seeking" phase, where the manifold's dimensionality expands substantially, coinciding with peak n-gram memorization. Subsequently, a "compression-seeking" phase imposes anisotropic consolidation, selectively preserving variance along dominant eigendirections while contracting others, a transition marked with significant improvement in downstream task performance. We show these phases can emerge from a fundamental interplay of cross-entropy optimization under skewed token frequencies and representational bottlenecks ($d \ll |V|$). Post-training further transforms geometry: SFT and DPO drive "entropy-seeking" dynamics to integrate specific instructional or preferential data, improving in-distribution performance while degrading out-of-distribution robustness. Conversely, RLVR induces "compression-seeking", enhancing reward alignment but reducing generation diversity.

A shifting graduate labour market is not unusual, said Kirsten Barnes, head of digital platform at Bright Network, which connects graduates and young professionals to employers. "Any shifts in the graduate job market this year – which typically fluctuates by 10-15% – appear to be driven by a combination of factors, including wider economic conditions and the usual fluctuations in business demand, rather than a direct impact from AI alone. We're not seeing a consistent trend across specific sectors," she said. Claire Tyler, head of insights at the Institute for Student Employers (ISE), which represents major graduate employers, said that among companies recruiting fewer graduates "none of them have said it's down to AI". Some recruitment specialists cited the recent increase in employer national insurance contributions as a factor in slowing down entry-level recruitment.

In this series of articles, we're publishing summaries with some of the key takeaways from a few of the workshops held at the 39th Annual AAAI Conference on Artificial Intelligence (AAAI 2025). Recent progress in Sphere Neural Networks demonstrates various possibilities for neural networks to achieve symbolic-level reasoning. This workshop aimed to reconsider various problems and discuss walk-round solutions in the two-way street commingling of neural networks and mathematics. This workshop brought together researchers from artificial intelligence and diverse scientific domains to address new challenges towards accelerating scientific discovery and engineering design. This was the fourth iteration of the workshop, with the theme of AI for biological sciences following previous three years' themes of AI for chemistry, earth sciences, and materials/manufacturing respectively.

One open question in the study of Large Language Models (LLMs) is whether they can emulate human ethical reasoning and act as believable proxies for human judgment. To investigate this, we introduce a benchmark dataset comprising 196 real-world ethical dilemmas and expert opinions, each segmented into five structured components: Introduction, Key Factors, Historical Theoretical Perspectives, Resolution Strategies, and Key Takeaways. We also collect non-expert human responses for comparison, limited to the Key Factors section due to their brevity. We evaluate multiple frontier LLMs (GPT-4o-mini, Claude-3.5-Sonnet, Deepseek-V3, Gemini-1.5-Flash) using a composite metric framework based on BLEU, Damerau-Levenshtein distance, TF-IDF cosine similarity, and Universal Sentence Encoder similarity. Metric weights are computed through an inversion-based ranking alignment and pairwise AHP analysis, enabling fine-grained comparison of model outputs to expert responses. Our results show that LLMs generally outperform non-expert humans in lexical and structural alignment, with GPT-4o-mini performing most consistently across all sections. However, all models struggle with historical grounding and proposing nuanced resolution strategies, which require contextual abstraction. Human responses, while less structured, occasionally achieve comparable semantic similarity, suggesting intuitive moral reasoning. These findings highlight both the strengths and current limitations of LLMs in ethical decision-making.

A speech by the US vice-president, JD Vance, symbolised a fracturing consensus on how to approach AI. He attended the summit with other global leaders, including the Indian prime minister, Narendra Modi, the Canadian PM, Justin Trudeau, and the head of the European Commission, Ursula von der Leyen. In his speech in the Grand Palais, Vance made it clear the US was not going to be held back from developing the tech by global regulation or an excessive focus on safety. "We need international regulatory regimes that foster the creation of AI technology rather than strangle it, and we need our European friends, in particular, to look to this new frontier with optimism rather than trepidation," he said. Speaking in front of the country's vice-premier, Zhang Guoqing, Vance warned his peers against cooperating with "authoritarian" regimes – in a clear reference to Beijing.

Sam Altman, the chief executive of ChatGPT's OpenAI, testified before members of a Senate subcommittee on Tuesday about the need to regulate the increasingly powerful artificial intelligence technology being created inside his company and others like Google and Microsoft. The three-hour-long hearing touched on several aspects of the risks that generative AI could pose to society, how it would affect the jobs market and why regulation by governments would be needed. Tuesday's hearing will be the first in a series of hearings to come as lawmakers grapple with drafting regulations around AI to address its ethical, legal and national security concerns. Senator Richard Blumenthal from Connecticut opened the proceedings with an AI-generated audio recording that sounded just like him. "Too often we have seen what happens when technology outpaces regulation. We have seen how algorithmic biases can perpetuate discrimination and prejudice and how the lack of transparency can undermine public trust. This is not the future we want," the voice said.

You've probably heard a lot about both Bing Chat and ChatGPT in recent weeks. Generative AI is here to stay and even if it's not something you can see yourself using in the long term, it's certainly worth trying these tools out and educating yourself a little more on them. Bing Chat is in some ways quite like ChatGPT. Microsoft is a big investor in OpenAI, the company behind ChatGPT, and uses the latest GPT-4 model in the backend of Bing Chat. So, there are definite similarities in the way the two operate.

The age of artificial intelligence is dawning. In contrast to AI's many benefits is the fact that it will displace millions of people around the world from their current workplace roles, especially those in white-collar jobs such as customer service, copywriting, and computer programming. It has already started to do so. Yet AI also presents a wonderful opportunity to rethink how we develop new skills. Those that seize this opportunity will move forward into exciting roles of their choosing, equipped with new skills they learned with the support of AI. To mitigate the negative impacts of AI on our careers, we must evolve our methods of acquiring new skills. In this post, I share my recent experience of interacting with ChatGPT while exploring the impact of permanently closing the United States stock exchanges.

Collaborative filtering is a clever recommendation system technique that predicts what you'll like based on others with similar tastes. Super useful for businesses like Netflix or Amazon to personalize suggestions! For example, if you and another user both love sci-fi, it'll recommend shows they enjoyed, knowing you'll probably like them as well. Learning latent factors is all about discovering hidden features that help explain user-item interactions, like why people like certain movies. For example, suppose we're recommending movies.