The international AI safety report warns systems are improving rapidly - but remain prone to'hallucinations' and hard to control. The international AI safety report warns systems are improving rapidly - but remain prone to'hallucinations' and hard to control. The International AI Safety report is an annual survey of technological progress and the risks it is creating across multiple areas, from deepfakes to the jobs market. Commissioned at the 2023 global AI safety summit, it is chaired by the Canadian computer scientist Yoshua Bengio, who describes the "daunting challenges" posed by rapid developments in the field. The report is also guided by senior advisers, including Nobel laureates Geoffrey Hinton and Daron Acemoglu.

Planning allows an agent to safely refine its actions before executing them in the real world. In autonomous driving, this is crucial to avoid collisions and navigate in complex, dense traffic scenarios. One way to plan is to search for the best action sequence. However, this is challenging when all necessary components - policy, next-state predictor, and critic - have to be learned. Here we propose Differentiable Simulation for Search (DSS), a framework that leverages the differentiable simulator Waymax as both a next state predictor and a critic. It relies on the simulator's hardcoded dynamics, making state predictions highly accurate, while utilizing the simulator's differentiability to effectively search across action sequences. Our DSS agent optimizes its actions using gradient descent over imagined future trajectories. We show experimentally that DSS - the combination of planning gradients and stochastic search - significantly improves tracking and path planning accuracy compared to sequence prediction, imitation learning, model-free RL, and other planning methods.

The big strides seen in generative AI are not based on somewhat obscure algorithms, but due to clearly defined generative principles. The resulting concrete implementations have proven themselves in large numbers of applications. We suggest that it is imperative to thoroughly investigate which of these generative principles may be operative also in the brain, and hence relevant for cognitive neuroscience. In addition, ML research led to a range of interesting characterizations of neural information processing systems. We discuss five examples, the shortcomings of world modelling, the generation of thought processes, attention, neural scaling laws, and quantization, that illustrate how much neuroscience could potentially learn from ML research.

How healthy am I? My immunome knows the score. Made up of 1.8 trillion cells and trillions more proteins, metabolites, mRNA, and other biomolecules, every person's immunome is different, and it is constantly changing. It's shaped by everything we have ever been exposed to physically and emotionally, and powerfully influences everything from our vulnerability to viruses and cancer to how well we age to whether we tolerate certain foods better than others. Yet as critical as the immunome is to each of us, it has remained largely beyond the reach of modern medicine. Now, thanks to a slew of new technologies, understanding this vital and mysterious system is within our grasp, paving the way for powerful new tools and tests to help us better assess, diagnose and treat diseases. On Monday, we published our 2025 edition of Climate Tech Companies to Watch .

What our latest list of Climate Tech Companies to Watch says about this moment. On Monday, we published our 2025 edition of Climate Tech Companies to Watch . This marks the third time we've put the list together, and it's become one of my favorite projects to work on every year. In the journalism world, it's easy to get caught up in the latest news, whether it's a fundraising round, research paper, or startup failure. Curating this list gives our team a chance to take a step back and consider the broader picture. What industries are making progress or lagging behind?

The next time you order a takeaway, it might be flown directly to your door. Today, Deliveroo has launched its first drone delivery service for customers in Ireland. Drones travelling at speeds of up to 50 miles per hour (80 kph) will carry food from restaurants to customers in as little as three minutes. Upon arrival, the drone will hover above the customer's home and gently lower the food to the ground on a tether before returning to the delivery hub. Launching in Blanchardstown, on the outskirts of Dublin, the trial will cover a 1.8-mile (3km) radius, reaching up to 150,000 people.

One key caveat here is that we don't know much about "closed source" models--for these, companies hold back the details of how they work. Instead, we worked with researchers who measured the energy it takes to run open-source AI models, for which the source code is publicly available. But using open-source models, it's possible to directly measure the energy used to respond to a query rather than just guess. We worked with researchers who generated text, images, and video and measured the energy required for the chips the models are based on to perform the task. Even just within the text responses, there was a pretty large range of energy needs.