Welcome back to, TIME's new twice-weekly newsletter about AI. If you're reading this in your browser, why not subscribe to have the next one delivered straight to your inbox? What to Know: Testing LLMs' ability to control a robot A couple of weeks ago, I wrote in this newsletter about my visit to Figure AI, a California startup that has developed a humanoid robot. Billions of dollars are currently pouring into the robotics industry, based on the belief that rapid AI progress will mean the creation of robots with "brains" that can finally deal with the messy complexities of the real world. Today, I want to tell you about an experiment that calls that theory into question.

An ex-Intel CEO's mission to build a Christian AI: 'hasten the coming of Christ's return' The Guardian's journalism is independent. We will earn a commission if you buy something through an affiliate link. I n March, three months after being forced out of his position as the CEO of Intel and sued by shareholders, Patrick Gelsinger took the reins at Gloo, a technology company made for what he calls the "faith ecosystem" - think Salesforce for churches, plus chatbots and AI assistants for automating pastoral work and ministry support. The Guardian's journalism is independent. We will earn a commission if you buy something through an affiliate link. The former CEO's career pivot is taking place as the US tech industry returns to the political realm as a major revenue stream.

The $226 million project uses ocean breezes and seawater to stay cool. China is submerging data centers into the ocean to keep them cool. China has completed the first phase of construction of what it claims is the world's first underwater data center (UDC). Located in Shanghai's Lin-gang Special Area with a price tag of roughly RMB 1.6 billion ($226 million), it's a significant milestone in the quest for sustainable solutions to the growing energy demands of China's computing infrastructure. Powered entirely by wind energy, the initiative has a total power capacity of 24 megawatts.

The 28th European Conference on Artificial Intelligence (ECAI-2025) is currently taking place in Bologna, Italy, running from 25-30 October 2025. During the opening ceremony, the winners of the ECAI-2025 and Prestigious Applications of Intelligent Systems (PAIS-2025) outstanding paper awards were announced. Letting AI agents interact in multi-agent applications adds a layer of complexity to the interpretability and prediction of AI outcomes, with profound implications for their trustworthy adoption in research and society. Game theory offers powerful models to capture and interpret strategic interaction among agents, but requires the support of reproducible, standardized and user-friendly IT frameworks to enable comparison and interpretation of results. We describe its implementation and usage, and we employ it to uncover biased outcomes in popular games among AI agents, depending on the employed Large Language Model (LLM) and used language, as well as on the personality trait or strategic knowledge of the agents.

Doctors are more like machines': my mother's worrying reliance on AI for health advice Tired of a two-day commute to see her overworked doctor, my mother turned to tech for help with her kidney disease. E very few months, my mother, a 57-year-old kidney transplant patient who lives in a small city in eastern China, embarks on a two-day journey to see her doctor. She fills her backpack with a change of clothes, a stack of medical reports and a few boiled eggs to snack on. Then, she takes a 90-minute ride on a high-speed train and checks into a hotel in the eastern metropolis of Hangzhou. At 7am the next day, she lines up with hundreds of others to get her blood taken in a long hospital hall that buzzes like a crowded marketplace. In the afternoon, when the lab results arrive, she makes her way to a specialist's clinic. She gets about three minutes with the doctor. Then, my mother packs up and starts the long commute home. My mother began using China's leading AI chatbot to diagnose her symptoms this past winter. She would lie down on her couch and open the app on her iPhone. "Hi," she said in her first message to the chatbot, on 2 February. How can I assist you today?" the system responded instantly, adding a smiley emoji.

BEIJING/SINGAPORE - Chinese state-owned defense giant Norinco in February unveiled a military vehicle capable of autonomously conducting combat-support operations at 50 kilometers per hour. It was powered by DeepSeek, the company whose artificial intelligence model is the pride of China's tech sector. The Norinco P60's release was touted by Communist Party officials in press statements as an early showcase of how Beijing is using DeepSeek and AI to catch up in its arms race with the United States, at a time when leaders in both countries have urged their militaries to prepare for conflict. A review of hundreds of research papers, patents and procurement records gives a snapshot of the systematic effort by Beijing to harness AI for military advantage. In a time of both misinformation and too much information, quality journalism is more crucial than ever.

The new AI-powered Wikipedia competitor falsely claims that pornography worsened the AIDS epidemic and that social media may be fueling a rise in transgender people. On Monday, Elon Musk's xAI startup launched Grokipedia, which the billionaire is pitching as an AI-generated alternative to the crowdsourced encyclopedia Wikipedia. Musk first announced the project in late September on his social media platform X, saying it would be "a massive improvement over Wikipedia," and "a necessary step towards the xAI goal of understanding the Universe." Musk said last week that he had delayed the launch of Grokipedia because his team needed "to do more work to purge out the propaganda." When Grokipedia eventually dropped on Monday, WIRED was initially unable to access the website and received an automated message that it was blocked.

This study proposes an interpretable prediction framework with literature-informed fine-tuned (LIFT) LLMs for truck driving risk prediction. The framework integrates an LLM-driven Inference Core that predicts and explains truck driving risk, a Literature Processing Pipeline that filters and summarizes domain-specific literature into a literature knowledge base, and a Result Evaluator that evaluates the prediction performance as well as the interpretability of the LIFT LLM. After fine-tuning on a real-world truck driving risk dataset, the LIFT LLM achieved accurate risk prediction, outperforming benchmark models by 26.7% in recall and 10.1% in F1-score. Furthermore, guided by the literature knowledge base automatically constructed from 299 domain papers, the LIFT LLM produced variable importance ranking consistent with that derived from the benchmark model, while demonstrating robustness in interpretation results to various data sampling conditions. The LIFT LLM also identified potential risky scenarios by detecting key combination of variables in truck driving risk, which were verified by PERMANOVA tests. Finally, we demonstrated the contribution of the literature knowledge base and the fine-tuning process in the interpretability of the LIFT LLM, and discussed the potential of the LIFT LLM in data-driven knowledge discovery.

Large Language Models' (LLMs) ability to converse naturally is empowered by their ability to empathetically understand and respond to their users. However, emotional experiences are shaped by demographic and cultural contexts. This raises an important question: Can LLMs demonstrate equitable empathy across diverse user groups? We propose a framework to investigate how LLMs' cognitive and affective empathy vary across user personas defined by intersecting demographic attributes. Our study introduces a novel intersectional analysis spanning 315 unique personas, constructed from combinations of age, culture, and gender, across four LLMs. Results show that attributes profoundly shape a model's empathetic responses. Interestingly, we see that adding multiple attributes at once can attenuate and reverse expected empathy patterns. We show that they broadly reflect real-world empathetic trends, with notable misalignments for certain groups, such as those from Confucian culture. We complement our quantitative findings with qualitative insights to uncover model behaviour patterns across different demographic groups. Our findings highlight the importance of designing empathy-aware LLMs that account for demographic diversity to promote more inclusive and equitable model behaviour.

Full fine-tuning of Large Language Models (LLMs) is notoriously memory-intensive, primarily because conventional optimizers such as SGD or Adam assume access to exact gradients derived from cached activations. Existing solutions either alter the model architecture (e.g., reversible networks) or trade memory for computation (e.g., activation checkpointing), but the optimizer itself remains untouched. In this work, we introduce GradLite, a backward-friendly optimizer that relaxes the requirement of exact gradients, enabling efficient training even when intermediate activations are aggressively discarded or approximated. GradLite leverages two key techniques: (i) low-rank Jacobian approximation, which reduces the dimensionality of backpropagated error signals, and (ii) error-feedback correction, which accumulates and compensates approximation errors across iterations to preserve convergence guarantees. We provide a theoretical analysis showing that GradLite maintains unbiased gradient estimates with bounded variance, ensuring convergence rates comparable to Adam. Empirically, GradLite reduces optimizer-state and activation memory consumption by up to 50\% without architectural changes, and achieves on-par or superior downstream performance on reasoning (MMLU, GSM8K), multilingual, and dialogue benchmarks compared to checkpointing and optimizer-centric baselines (LoMo, GaLore).