Global climate change has reduced crop resilience and pesticide efficacy, making reliance on synthetic pesticides inevitable, even though their widespread use poses significant health and environmental risks. While these pesticides remain a key tool in pest management, previous machine-learning applications in pesticide and agriculture have focused on classification or regression, leaving the fundamental challenge of generating new molecular structures or designing novel candidates unaddressed. In this paper, we propose Pesti-Gen, a novel generative model based on variational auto-encoders, designed to create pesticide candidates with optimized properties for the first time. Specifically, Pesti-Gen leverages a two-stage learning process: an initial pre-training phase that captures a generalized chemical structure representation, followed by a fine-tuning stage that incorporates toxicity-specific information. The model simultaneously optimizes over multiple toxicity metrics, such as (1) livestock toxicity and (2) aqua toxicity to generate environmentally friendly pesticide candidates. Notably, Pesti-Gen achieves approximately 68\% structural validity in generating new molecular structures, demonstrating the model's effectiveness in producing optimized and feasible pesticide candidates, thereby providing a new way for safer and more sustainable pest management solutions.

We present symbol tuning - finetuning language models on in-context input-label pairs where natural language labels (e.g., "positive/negative sentiment") are replaced with arbitrary symbols (e.g., "foo/bar"). Symbol tuning leverages the intuition that when a model cannot use instructions or natural language labels to figure out a task, it must instead do so by learning the input-label mappings. We experiment with symbol tuning across Flan-PaLM models up to 540B parameters and observe benefits across various settings. First, symbol tuning boosts performance on unseen in-context learning tasks and is much more robust to underspecified prompts, such as those without instructions or without natural language labels. Second, symbol-tuned models are much stronger at algorithmic reasoning tasks, with up to 18.2% better performance on the List Functions benchmark and up to 15.3% better performance on the Simple Turing Concepts benchmark. Finally, symbol-tuned models show large improvements in following flipped-labels presented in-context, meaning that they are more capable of using in-context information to override prior semantic knowledge.

From public CCTV cameras to biometric identification systems in airports, facial recognition technology is now common in a growing number of places around the world. In its most benign form, facial recognition technology is a convenient way to unlock your smartphone. At the state level though, facial recognition is a key component of mass surveillance, and it already touches half the global population on a regular basis. Today's visualizations from SurfShark classify 194 countries and regions based on the extent of surveillance. Click here to explore the full research methodology.

How many trees are in your city? It might seem like a straightforward question, but finding the answer can be a monumental task. New York City's 2015-2016 tree census, for example, took nearly two years (12,000 hours total) and more than 2,200 volunteers. Seattle's tree inventory won't be complete until at least 2024. Such efforts aren't done in vain; in the short term, they allow cities to better maintain their urban trees.

This project demonstrated how modern data science can greatly increase the efficiency of conventional research at sea, and improve the productivity of interactive seafloor exploration with the all too familiar "stumbling in the dark" mode. "Developing totally new operational workflows is risky, however, it is very relevant for applications such as seafloor monitoring, ecosystem survey and planning the installation and decommissioning of seafloor infrastructure," said Thornton. The idea behind this Adaptive Robotics mission was not to upturn the structure of how things are done at sea, but simply to remove bottlenecks in the flow of information and data-processing using computational methods and Artificial Intelligence. The algorithms used are able to rapidly produce simple summaries of observations, and form subsequent deployment plans. This way, scientists can respond to dynamic changes in the environment and target areas that will lead to the biggest operational, scientific, or environmental management gains.

Kathryn Hess can't tell the difference between a coffee mug and a bagel. Hess, a researcher at the Swiss Federal Institute of Technology, is one of the world's leading thinkers in the field of algebraic topology--in super simplified terms, the mathematics of rubbery shapes. It uses algebra to attack the following question: If given two geometric objects, can you deform one to another without making any cuts? The answer, when it comes to bagels and coffee mugs, is yes, yes you can. If that all sounds annoyingly abstract, well, it kind of is.

Forget dot coms and social networks. The hotspot for research and investment in Silicon Valley right now is the future of transport. Convince the valley you have a new way to create a brain for a self-driving car, help people find parking, detect a drowsy driver, or build a personal electric plane, and you'll find yourself showered in VC funding. That explains the madness of the above chart (desktop users: mouse over for a zoomed in view). The "Future of Transportation Stack," produced by VC firm Comet Labs, counts 263 companies, most of which you've probably never heard of, all of them vying to cash in on the nascent automotive revolution.