The Download: climate tech companies to watch and AI's discovery problem Plus: OpenAI has scrapped a new AI model over safety concerns. With the planet nearing 1.5 C of warming, climate policies being unraveled, and Big Tech backpedaling on its climate ambitions, it can be tempting to give in to climate doom and defeatism. But despite the headwinds, the world has still made incredible progress. That's why publishes its annual list of Climate Tech Companies to Watch . Once a year, we compile a list of 10 companies that we believe have done the most, or have the best chance, to make a real dent in emissions or improve public safety and health. We've now finalized our 2026 list and will publish it on October 6.

When can we say AI made a scientific discovery? AI companies' insistence that their technology is making breakthroughs, not just aiding scientists, is making real progress harder to recognize. Last Wednesday, Anthropic announced that earlier this year it had launched a molecular biology lab, where Claude agents read and conjecture about hard biology problems and human scientists run experiments on what they report. And this AI-powered lab, the company said, had made its first discovery. To understand what Anthropic says its system did, imagine you're flipping through a library of millions of DNA sequences, amassed as scientists sequence more and more of the living world. One step toward a breakthrough might be finding a peculiar sequence that encodes an interesting enzyme, perhaps.

The potential for AI to automate scientific research and manufacturing must be balanced with new risks, Anthropic says. Artificial intelligence agents might occasionally get confused and hack into other computers, but Anthropic thinks it has a way to unleash the little rascals into scientific labs and manufacturing facilities safely. The AI company released details today of a new framework designed to help AI agents use physical systems like microscopes, liquid-handling equipment, quantum computing hardware, manufacturing machines, and robot arms. The framework, called Model Hardware Standard, is a set of rules that specify how AI agents should--and should not--interact with all sorts of hardware. It reflects a growing belief that AI has the potential to revolutionize scientific research and industries like manufacturing-if it can venture into the physical world safely.

Gateway spaceport will make Earth's air traffic control look easy. More information Adding us as a Preferred Source in Google by using this link indicates that you would like to see more of our content in Google News results. The Gateway space station will be humanity's first space station around the Moon as a vital component of the Artemis missions to return humans to the lunar surface for scientific discovery and chart the path for the first human missions to Mars. Astronauts on Gateway will be the first humans to call deep space home during missions where they will use Gateway to conduct science and prepare for lunar surface missions. Breakthroughs, discoveries, and DIY tips sent six days a week.

Akari describes her work on augmented language models, where language models are trained to use other models and tools. This has already resulted in OpenScholar, an open-source model which helps scientists to synthesize vast amounts of scientific literature. You were awarded the 2025 AAAI Doctoral Dissertation Award. What was the topic of your dissertation research, and why was this an interesting area of study to you? My PhD dissertation was on Retrieval-Augmented Language Models.

Plus: A new Amazon data center could become the US's most polluting power plant. In 2024, Google DeepMind scientists shared the Nobel Prize in Chemistry for a neural network, AlphaFold, which predicts the structures of proteins. It showed that AI could make groundbreaking scientific discoveries, but AlphaFold may not be the best template for accelerating science. Instead, another approach may hold the key: AI agents. AlphaFold relied on a dataset of roughly 170,000 experimentally validated protein structures that took 53 years and roughly $21 billion worth of experimental work to assemble. Comparable datasets will be difficult or impossible to create in many fields.

Benjamin Franklin and other citizen scientists are core parts of the American experiment. More information Adding us as a Preferred Source in Google by using this link indicates that you would like to see more of our content in Google News results. Benjamin Franklin's kite experiment in 1752 was a pivotal scientific event, which demonstrated the connection between lightning and electricity. Breakthroughs, discoveries, and DIY tips sent six days a week. By signing up, you confirm you are 16+, will receive newsletters and promotional content and agree to our Terms of Use and acknowledge the data practices in our Privacy Policy .

Scientific discovery via symbolic regression is often viewed as statistically and computationally intractable because the hypothesis space of expressions grows combinatorially with depth. This paper revisits the statistical side through the lens of PAC learning, focusing on compositional function trees built from a finite vocabulary of smooth operators (e.g., $\{+,\times,\sin,\exp\}$ and affine maps). We prove that the relevant generalization quantity, Rademacher complexity, hence the excess risk, does not necessarily blow up exponentially with the number of distinct symbolic structures, but is controlled by (i) the depth $d$ and (ii) the Lipschitz constants of the base operators along the composed computation graph. Concretely, under mild Lipschitz conditions on operators and bounded affine leaves, a finite-union bound over a vocabulary of size $K=|\mathcal{H}_{\mathrm{base}}|$ together with Maurer-type vector contraction yields $\mathfrak{R}_n(\mathcal{H}_{\mathrm{comp}}^{d}) \leq (Kb\sqrt{2}L)^{d-1}\mathfrak{R}_n(\mathcal{H}_{\mathrm{comp}}^{1})$ with arity bound $b$; corresponding high-probability risk bounds scale as $\mathcal{O}(L^{d}/\sqrt{n})$ when $K,b=O(1)$ and $\mathfrak{R}_n(\mathcal{H}_{\mathrm{comp}}^{1})=O(n^{-1/2})$. We complement the theory with a modular codebase that trains differentiable operator trees (not MLPs) on synthetic "physics-like" targets of controlled depth and shows that the empirical generalization gap correlates positively with the predicted complexity term $(\widehat{L}^{d})/\sqrt{n}$.

Why are airplanes so cold? From combating fainting to helping aircraft work efficiently, planes are chilly for a reason. More information Adding us as a Preferred Source in Google by using this link indicates that you would like to see more of our content in Google News results. Breakthroughs, discoveries, and DIY tips sent six days a week. By signing up, you confirm you are 16+, will receive newsletters and promotional content and agree to our Terms of Use and acknowledge the data practices in our Privacy Policy .

We examine hypothesis testing within a principal-agent framework, where a strategic agent, holding private beliefs about the effectiveness of a product, submits data to a principal who decides on approval. The principal employs a hypothesis testing rule, aiming to pick a p-value threshold that balances false positives and false negatives while anticipating the agent's incentive to maximize expected profitability. Building on prior work, we develop a game-theoretic model that captures how the agent's participation and reporting behavior respond to the principal's statistical decision rule. Despite the complexity of the interaction, we show that the principal's errors exhibit clear monotonic behavior when segmented by an efficiently computable critical p-value threshold, leading to an interpretable characterization of their optimal p-value threshold.