We present a geometric framework for understanding Transformer-based language models, drawing an explicit analogy to General Relativity. Queries and keys induce an effective metric on representation space, and attention acts as a discrete connection that implements parallel transport of value vectors across tokens. Stacked layers provide discrete time-slices through which token representations evolve on this curved manifold, while backpropagation plays the role of a least-action principle that shapes loss-minimizing trajectories in parameter space. If this analogy is correct, token embeddings should not traverse straight paths in feature space; instead, their layer-wise steps should bend and reorient as interactions mediated by embedding space curvature. To test this prediction, we design experiments that expose both the presence and the consequences of curvature: (i) we visualize a curvature landscape for a full paragraph, revealing how local turning angles vary across tokens and layers; (ii) we show through simulations that excess counts of sharp/flat angles and longer length-to-chord ratios are not explainable by dimensionality or chance; and (iii) inspired by Einstein's eclipse experiment, we probe deflection under controlled context edits, demonstrating measurable, meaning-consistent bends in embedding trajectories that confirm attention-induced curvature.

Knowledge editing and machine unlearning are two popular approaches for large language models (LLMs) to stay up-to-date. However, the knowledge updating mechanism of LLMs remains largely unexplored due to insufficient, isolated, and small-scale evaluation. For instance, are LLMs similar to humans in modifying certain knowledge? What differs editing and unlearning as training data increases? This paper proposes KnowledgeSmith, a unified framework to systematically understand the updating mechanism of LLMs. We first cast editing and unlearning as instances of one constrained optimization problem. Then, we propose an automatic dataset generator that provides structured interventions across multiple graph levels and data scales, enabling controlled studies of how different modification strategies propagate through model knowledge. Extensive experiments demonstrate nuanced insights over knowledge propagation, plasticity scaling, consistency, and robustness. For instance, our results show that LLMs do not exhibit similar updating as humans for different levels of knowledge, and there exists consistency-capacity trade-off. We hope our findings can offer suggestions to the design of more reliable and scalable strategies. Code: https://github.com/AIFrontierLab/KnowledgeSmith.git

We present GravLensX, an innovative method for rendering black holes with gravitational lensing effects using neural networks. The methodology involves training neural networks to fit the spacetime around black holes and then employing these trained models to generate the path of light rays affected by gravitational lensing. This enables efficient and scalable simulations of black holes with optically thin accretion disks, significantly decreasing the time required for rendering compared to traditional methods. We validate our approach through extensive rendering of multiple black hole systems with superposed Kerr metric, demonstrating its capability to produce accurate visualizations with significantly $15\times$ reduced computational time. Our findings suggest that neural networks offer a promising alternative for rendering complex astrophysical phenomena, potentially paving a new path to astronomical visualization.

There are new hints that the fabric of space-time may be made of "memory cells" that record the whole history of the universe. If true, it could explain the nature of dark matter and much more

Compute optimization using token reduction of LLM prompts is an emerging task in the fields of NLP and next generation, agentic AI. In this white paper, we introduce a novel (patent pending) text representation scheme and a first-of-its-kind word-level semantic compression of paragraphs that can lead to over 90% token reduction, while retaining high semantic similarity to the source text. We explain how this novel compression technique can be lossless and how the detail granularity is controllable. We discuss benchmark results over open source data (i.e. Bram Stoker's Dracula available through Project Gutenberg) and show how our results hold at the paragraph level, across multiple genres and models.

The Trump Administration quietly revealed it has futuristic technologies that literally bend time during a speech on'the golden age of American innovation.' The director of the White House Office of Science and Technology Policy, Michael Kratsios, declared that the US currently has the ability to'manipulate time and space' and'leave distance annihilated.' Kratsios made the bold statement on Monday during the Endless Frontiers Retreat, a scientific conference in Texas focused on promoting US technological innovations to maintain global competitiveness. The rest of the director's speech touched on American breakthroughs of the past and undoing Biden-era policies that the Trump Administration claims stifled innovation - adding that the regulatory process on new tech has been a burden since the 1970s. Kratsios actually referenced this again at the end of his speech, saying that Americans will soon have the choice to'craft new technologies and give themselves to scientific discoveries that will bend time and space.'

From H. G. Wells's The Time Machine to Christopher Nolan's Interstellar, the possibility of travelling through time has fascinated people for centuries. But, although it sounds like pure science fiction, physicists now believe that time travel really is possible. In fact, scientists say that people have already done it. But, before you start to plan your trip to ancient Rome, the experts caution that real time travel is nothing like what you see in the movies. It might seem obvious, but here on Earth, we all move through time at a speed of one second per second.

We explore local vs. global evolution of knowledge systems through the framework of socio-epistemic networks (SEN), applying two complementary methods to a corpus of scientific texts. The framework comprises three interconnected layers-social, semiotic (material), and semantic-proposing a multilayered approach to understanding structural developments of knowledge. To analyse diachronic changes on the semantic layer, we first use information-theoretic measures based on relative entropy to detect semantic shifts, assess their significance, and identify key driving features. Second, variations in document embedding densities reveal changes in semantic neighbourhoods, tracking how concentration of similar documents increase, remain stable, or disperse. This enables us to trace document trajectories based on content (topics) or metadata (authorship, institution). Case studies of Joseph Silk and Hans-J\"urgen Treder illustrate how individual scholar's work aligns with broader disciplinary shifts in general relativity and gravitation research, demonstrating the applications, limitations, and further potential of this approach.

Generating high-resolution simulations is key for advancing our understanding of one of the universe's most violent events: Black Hole mergers. However, generating Black Hole simulations is limited by prohibitive computational costs and scalability issues, reducing the simulation's fidelity and resolution achievable within reasonable time frames and resources. In this work, we introduce a novel method that circumvents these limitations by applying a super-resolution technique without directly needing high-resolution labels, leveraging the Hamiltonian and momentum constraints--fundamental equations in general relativity that govern the dynamics of spacetime. We demonstrate that our method achieves a reduction in constraint violation by one to two orders of magnitude and generalizes effectively to out-of-distribution simulations.

While many of these interactions happen through direct contact, like the concert-goers' nudging, some interactions can transmit through the material the objects are on or in -- these are known as indirect interactions. For example, a bridge with pedestrians on it can transmit vibrations, like in the famous Millennium Bridge "wobbly bridge" instance. While the results of direct interactions (like nudging) are of increasing interest and study, and the results of indirect interactions through mechanisms like vision are well-studied, researchers are still learning about indirect mechanical interactions (for example, how two rolling balls might influence each other's movement on a trampoline by indenting the trampoline's surface with their weight, thus exerting mechanical forces without touching). Physicists are using small wheeled robots to better understand these indirect mechanical interactions, how they play a role in active matter, and how we can control them. Their findings, "Field-mediated locomotor dynamics on highly deformable surfaces" are recently published in the The Proceedings of the National Academy of Sciences (PNAS).