Newest Papers
9 projects i’m working on that makes me forget i own a phone
candice
Substack·2025
you simply have to become too busy to scroll
No ratings yet
View paper →Searching for correlations between satellite galaxy populations and the cold circumgalactic medium around TNG50 galaxies
Mitali Damle, Stephanie Tonnesen, Martin Sparre, Philipp Richter
arXiv·2025
We investigate the impact of satellites, a potentially important contributor towards the cold gas assembly of a halo, on the cold gas budgets of 197 TNG50 simulated halos with masses of 10$^{10.85}$ $\le$ M$_{200c}$/M$_{\odot}$ $\le$ 10$^{12.24}$ at $z$ = 0. To highlight the effect of satellites, we split the sample into three mass bins. We find that the total number of satellites, total mass of satellites, number of massive satellites and stellar mass of the most massive satellite, all correlate with the cold gas mass in halos. The total number of satellites (stellar mass of the most massive satellite) correlates most with the halo cold gas mass for low (middle) mass halos. The number of massive or observable satellites correlates with cold gas mass in similar manner as the total number of satellites. Our findings can, therefore, be used to guide future observers to focus on the link between the number of observable satellites and the amount of cold gas in a halo. Despite this correlation, we find that much of the cold gas lies far from the satellites. This leads us to conclude that satellites are unlikely to be the main supplier for cold gas in halos, however we discuss how they may act in tandem with other sources such that the satellite population correlates with the total cold gas in their host halo.
★ 3.0 (1)
View paper →Effective context engineering for AI agents
Anthropic
Anthropic Blog·2025
Anthropic is an AI safety and research company that's working to build reliable, interpretable, and steerable AI systems.
No ratings yet
View paper →The Return of Great-Power Diplomacy | Foreign Affairs
A. Wess Mitchell
Foreign Affairs·2025
How strategic dealmaking can fortify American power.
No ratings yet
View paper →Trends in AI Supercomputers
Konstantin F. Pilz, James Sanders, Robi Rahman, Lennart Heim
arXiv·2025
Frontier AI development relies on powerful AI supercomputers, yet analysis of these systems is limited. We create a dataset of 500 AI supercomputers from 2019 to 2025 and analyze key trends in performance, power needs, hardware cost, ownership, and global distribution. We find that the computational performance of AI supercomputers has doubled every nine months, while hardware acquisition cost and power needs both doubled every year. The leading system in March 2025, xAI's Colossus, used 200,000 AI chips, had a hardware cost of $7B, and required 300 MW of power, as much as 250,000 households. As AI supercomputers evolved from tools for science to industrial machines, companies rapidly expanded their share of total AI supercomputer performance, while the share of governments and academia diminished. Globally, the United States accounts for about 75% of total performance in our dataset, with China in second place at 15%. If the observed trends continue, the leading AI supercomputer in 2030 will achieve 16-bit FLOP/s, use two million AI chips, have a hardware cost of $200 billion, and require 9 GW of power. Our analysis provides visibility into the AI supercomputer landscape, allowing policymakers to assess key AI trends like resource needs, ownership, and national competitiveness.
No ratings yet
View paper →The (Limited) Effect of Viscosity in Multiphase Turbulent Mixing
Tirso Marin-Gilabert, Max Gronke, S. Peng Oh
arXiv·2025
Multiphase gas can be found in many astrophysical environments, such as galactic outflows, stellar wind bubbles, and the circumgalactic medium, where the interplay between turbulence, cooling, and viscosity can significantly influence gas dynamics and star formation processes. We investigate the role of viscosity in modulating turbulence and radiative cooling in turbulent radiative mixing layers (TRMLs). In particular, we aim to determine how different amounts of viscosity affect the Kelvin-Helmholtz instability (KHI), turbulence evolution, and the efficiency of gas mixing and cooling. Using idealized 2D numerical setups, we compute the critical viscosity required to suppress the KHI in shear flows characterized by different density contrasts and Mach numbers. These results are then used in a 3D shear layer setup to explore the impact of viscosity on cooling efficiency and turbulence across different cooling regimes. We find that the critical viscosity follows the expected dependence on overdensity and Mach number. Our viscous TRMLs simulations show different behaviors in the weak and strong cooling regimes. In the weak cooling regime, viscosity has a strong impact, resulting in laminar flows and breaking previously established inviscid relations between cooling and turbulence (albeit leaving the total luminosity unaffected). However, in the strong cooling regime, when cooling timescales are shorter than viscous timescales, key scaling relations in TRMLs remain largely intact. In this regime -- which must hold for gas to remain multiphase -- radiative losses dominate, and the system effectively behaves as non-viscous regardless of the actual level of viscosity. Our findings have direct implications for both the interpretation of observational diagnostics and the development of subgrid models in large-scale simulations.
No ratings yet
View paper →Expected Free Energy-based Planning as Variational Inference
Bert de Vries, Wouter Nuijten, Thijs van de Laar, Wouter Kouw, Sepideh Adamiat, Tim Nisslbeck, Mykola Lukashchuk, Hoang Minh Huu Nguyen, Marco Hidalgo Araya, Raphael Tresor, Thijs Jenneskens, Ivana Nikoloska, Raaja Ganapathy Subramanian, Bart van Erp, Dmitry Bagaev, Albert Podusenko
arXiv·2025
We address the problem of planning under uncertainty, where an agent must choose actions that not only achieve desired outcomes but also reduce uncertainty. Traditional methods often treat exploration and exploitation as separate objectives, lacking a unified inferential foundation. Active inference, grounded in the Free Energy Principle, provides such a foundation by minimizing Expected Free Energy (EFE), a cost function that combines utility with epistemic drives, such as ambiguity resolution and novelty seeking. However, the computational burden of EFE minimization had remained a significant obstacle to its scalability. In this paper, we show that EFE-based planning arises naturally from minimizing a variational free energy functional on a generative model augmented with preference and epistemic priors. This result reinforces theoretical consistency with the Free Energy Principle by casting planning under uncertainty itself as a form of variational inference. Our formulation yields policies that jointly support goal achievement and information gain, while incorporating a complexity term that accounts for bounded computational resources. This unifying framework connects and extends existing methods, enabling scalable, resource-aware implementations of active inference agents.
No ratings yet
View paper →Achieving nature-human harmony: awe brings benevolence through nature connectedness
Silin Lin, Xiaolan Chen, Liying Bai
Current Psychology·2025·2 citations
No abstract available
No ratings yet
View paper →Singlet-Triplet Inversion
Liam Wrigley, Cody W. Schlenker
Annual Review of Physical Chemistry·2025·4 citations
<jats:p>The inversion of singlet and triplet states is a rare phenomenon, where, in opposition to Hund's first rule, singlet electronic states are stabilized relative to their triplet counterparts. The recent discovery of organic molecules exhibiting this inversion presents exciting new technological opportunities, such as addressing stability issues in organic light-emitting diodes (OLEDs). In this review, we describe fundamental molecular properties that can yield singlet-triplet inversion, generally ascribed to a phenomenon known as dynamic spin polarization. We discuss the systems in which singlet-triplet inversion was theoretically proposed, experimentally verified, and first implemented in an OLED device. We highlight key insights from the extensive computational work being carried out to understand the intricacies of these systems. Finally, we consider the outlook for future inverted singlet-triplet (IST) emitters.</jats:p>
No ratings yet
View paper →Does Reinforcement Learning Really Incentivize Reasoning Capacity in LLMs Beyond the Base Model?
Yang Yue, Zhiqi Chen, Rui Lu, Andrew Zhao, Zhaokai Wang, Shiji Song, Gao Huang
arXiv·2025
Reinforcement Learning with Verifiable Rewards (RLVR) has recently demonstrated notable success in enhancing the reasoning performance of large language models (LLMs), particularly on mathematics and programming tasks. Similar to how traditional RL helps agents explore and learn new strategies, RLVR is believed to enable LLMs to continuously self-improve, thus acquiring novel reasoning abilities beyond those of the corresponding base models. In this study we critically examine the current state of RLVR by systematically probing the reasoning capability boundaries of RLVR-trained LLMs across various model families, RL algorithms, and math, coding, and visual reasoning benchmarks, using pass@k at large k values as the evaluation metric. Surprisingly, we find that the current training setup does not elicit fundamentally new reasoning patterns. While RLVR-trained models outperform their base models at small k (e.g., k = 1), the base models achieve a higher pass@k score when k is large. Coverage and perplexity analyses show that the observed reasoning abilities originate from and are bounded by the base model. Treating the base model as an upper bound, our quantitative analysis shows that six popular RLVR algorithms perform similarly and remain far from optimal in leveraging the potential of the base model. By contrast, we find that distillation can introduce new reasoning patterns from the teacher and genuinely expand the model's reasoning capabilities. Overall, our findings suggest that current RLVR methods have not yet realized the potential of RL to elicit truly novel reasoning abilities in LLMs. This highlights the need for improved RL paradigms, such as continual scaling and multi-turn agent-environment interaction, to unlock this potential.
No ratings yet
View paper →Not All Rollouts are Useful: Down-Sampling Rollouts in LLM Reinforcement Learning
Yixuan Even Xu, Yash Savani, Fei Fang, J. Zico Kolter
arXiv·2025
Reinforcement learning with verifiable rewards (RLVR) has emerged as the leading approach for enhancing reasoning capabilities in large language models. However, it faces a fundamental compute and memory asymmetry: rollout generation is embarrassingly parallel and memory-light, whereas policy updates are communication-heavy and memory-intensive. To address this, we introduce PODS (Policy Optimization with Down-Sampling), which decouples rollout generation from policy updates by training only on a strategically selected subset of rollouts, maintaining learning quality while dramatically reducing update costs. We propose a principled subset selection criterion, max-variance down-sampling, that maximizes reward diversity, and provide an efficient $O(n\log n)$ implementation. Empirically, Group Relative Policy Optimization (GRPO) with PODS achieves the peak test accuracy of vanilla GRPO at least $\mathbf{1.7\times}$ faster across the different reasoning benchmarks and hardware configurations we tested.
No ratings yet
View paper →Artificial intelligence-assisted academic writing: recommendations for ethical use
Adam Cheng, Aaron Calhoun, Gabriel Reedy
Advances in Simulation·2025·37 citations
<jats:title>Abstract</jats:title>
<jats:p>Generative artificial intelligence (AI) tools have been selectively adopted across the academic community to help researchers complete tasks in a more efficient manner. The widespread release of the Chat Generative Pre-trained Transformer (ChatGPT) platform in 2022 has made these tools more accessible to scholars around the world. Despite their tremendous potential, studies have uncovered that large language model (LLM)-based generative AI tools have issues with plagiarism, AI hallucinations, and inaccurate or fabricated references. This raises legitimate concern about the utility, accuracy, and integrity of AI when used to write academic manuscripts. Currently, there is little clear guidance for healthcare simulation scholars outlining the ways that generative AI could be used to legitimately support the production of academic literature. In this paper, we discuss how widely available, LLM-powered generative AI tools (e.g. ChatGPT) can help in the academic writing process. We first explore how academic publishers are positioning the use of generative AI tools and then describe potential issues with using these tools in the academic writing process. Finally, we discuss three categories of specific ways generative AI tools can be used in an ethically sound manner and offer four key principles that can help guide researchers to produce high-quality research outputs with the highest of academic integrity.</jats:p>
No ratings yet
View paper →