Large Language Models (LLMs) are increasingly deployed in multi-agent systems, where effective inter-model communication is crucial. Existing communication protocols either rely on natural language, incurring high inference costs and information loss, or on hidden states, which suffer from information concentration bias and inefficiency. To address these limitations, we propose KVComm, a novel communication framework that enables efficient communication between LLMs through selective sharing of KV pairs. KVComm leverages the rich information encoded in the KV pairs while avoiding the pitfalls of hidden states. We introduce a KV layer-wise selection strategy based on attention importance scores with a Gaussian prior to identify the most informative KV pairs for communication. Extensive experiments across diverse tasks and model pairs demonstrate that KVComm achieves comparable performance to the upper-bound method, which directly merges inputs to one model without any communication, while transmitting as few as 30% of layers' KV pairs. Our study highlights the potential of KV pairs as an effective medium for inter-LLM communication, paving the way for scalable and efficient multi-agent systems. Large Language Models (LLMs) have catalyzed a paradigm shift from isolated model capabilities towards collaborative multi-agent systems (Guo et al., 2024; Tran et al., 2025). CAMEL (Li et al., 2023), AutoGen (Wu et al., 2024), and ChatDev (Qian et al., 2023) have demonstrated the potential of LLMs to collaborate effectively in multi-agent systems, achieving impressive results in various tasks. These systems leverage the strengths of individual LLMs and enable them to work together to solve complex problems that are beyond the capabilities of a single model (Y ang et al., 2024a).

Preparing high-quality datasets required by various data-driven AI and machine learning models has become a cornerstone task in data-driven analysis. Conventional data discovery methods typically integrate datasets towards a single pre-defined quality measure that may lead to bias for downstream tasks. This paper introduces MODis, a framework that discovers datasets by optimizing multiple user-defined, model-performance measures. Given a set of data sources and a model, MODis selects and integrates data sources into a skyline dataset, over which the model is expected to have the desired performance in all the performance measures. We formulate MODis as a multi-goal finite state transducer, and derive three feasible algorithms to generate skyline datasets. Our first algorithm adopts a "reduce-from-universal" strategy, that starts with a universal schema and iteratively prunes unpromising data. Our second algorithm further reduces the cost with a bi-directional strategy that interleaves data augmentation and reduction. We also introduce a diversification algorithm to mitigate the bias in skyline datasets. We experimentally verify the efficiency and effectiveness of our skyline data discovery algorithms, and showcase their applications in optimizing data science pipelines.

A recent influx of videos supposedly showing "drones" or other spooky unidentified aerial phenomena flying over darkened US skylines appears to be the result, in part, of AI-trickery. Since late November, residents in New Jersey and at least five other states have reported spotting bright objects flying overhead. The sightings have stirred speculation, amplified by celebrities, commentators, and prominent public officials, that this is nefarious, experimental technology. Now, several of the viral videos surfacing on TikTok and X over the past week are capitalizing on the panic; they also appear to exhibit the hallmark calling cards of generative AI manipulation. Almost none of the videos reviewed by Popular Science had any official label or disclosure from social media platforms warning users about possible digital editing.

A popular new method in mechanistic interpretability is to train high-dimensional sparse autoencoders (SAEs) on neuron activations and use SAE features as the atomic units of analysis. However, the body of evidence on whether SAE feature spaces are useful for causal analysis is underdeveloped. In this work, we use the RAVEL benchmark to evaluate whether SAEs trained on hidden representations of GPT-2 small have sets of features that separately mediate knowledge of which country a city is in and which continent it is in. We evaluate four open-source SAEs for GPT-2 small against each other, with neurons serving as a baseline, and linear features learned via distributed alignment search (DAS) serving as a skyline. For each, we learn a binary mask to select features that will be patched to change the country of a city without changing the continent, or vice versa. Our results show that SAEs struggle to reach the neuron baseline, and none come close to the DAS skyline. We release code here: https://github.com/MaheepChaudhary/SAE-Ravel

Tourists and workers alike jostling their way through New York's bustling midtown may notice an odd sight next time they look up. Dozens of floors above ground, the world's first commercial window-cleaning-robot will be thrusting its two white mechanical arms back and forth, soapy squeegees in hand. Skyline Robotics, the New York-based company behind the "Ozmo" cleaning robot, believe machines like theirs are faster and safer than traditional cleaning methods and could help address the potential shortage of human skyscraper window washers in coming years. It's just the latest example of artificial intelligence and robotics merging together to perform real-word tasks once confined to people. Starting this week, Skyline's Ozmo robot will get to work cleaning windows at 1133 Avenue of the Americas, a 45-story Class A skyscraper owned and managed by the Durst Organization near New York's Bryant Park.

In this paper, we present a Computer Vision (CV) based tracking and fusion algorithm, dedicated to a 3D printed gimbal system on drones operating in nature. The whole gimbal system can stabilize the camera orientation robustly in a challenging nature scenario by using skyline and ground plane as references. Our main contributions are the following: a) a light-weight Resnet-18 backbone network model was trained from scratch, and deployed onto the Jetson Nano platform to segment the image into binary parts (ground and sky); b) our geometry assumption from nature cues delivers the potential for robust visual tracking by using the skyline and ground plane as a reference; c) a spherical surface-based adaptive particle sampling, can fuse orientation from multiple sensor sources flexibly. The whole algorithm pipeline is tested on our customized gimbal module including Jetson and other hardware components. The experiments were performed on top of a building in the real landscape.

Training deep networks and tuning hyperparameters on large datasets is computationally intensive. One of the primary research directions for efficient training is to reduce training costs by selecting well-generalizable subsets of training data. Compared to simple adaptive random subset selection baselines, existing intelligent subset selection approaches are not competitive due to the time-consuming subset selection step, which involves computing model-dependent gradients and feature embeddings and applies greedy maximization of submodular objectives. Our key insight is that removing the reliance on downstream model parameters enables subset selection as a pre-processing step and enables one to train multiple models at no additional cost. In this work, we propose MILO, a model-agnostic subset selection framework that decouples the subset selection from model training while enabling superior model convergence and performance by using an easy-to-hard curriculum. Our empirical results indicate that MILO can train models $3\times - 10 \times$ faster and tune hyperparameters $20\times - 75 \times$ faster than full-dataset training or tuning without compromising performance.

Radial basis function neural networks (\emph{RBFNN}) are {well-known} for their capability to approximate any continuous function on a closed bounded set with arbitrary precision given enough hidden neurons. In this paper, we introduce the first algorithm to construct coresets for \emph{RBFNNs}, i.e., small weighted subsets that approximate the loss of the input data on any radial basis function network and thus approximate any function defined by an \emph{RBFNN} on the larger input data. In particular, we construct coresets for radial basis and Laplacian loss functions. We then use our coresets to obtain a provable data subset selection algorithm for training deep neural networks. Since our coresets approximate every function, they also approximate the gradient of each weight in a neural network, which is a particular function on the input. We then perform empirical evaluations on function approximation and dataset subset selection on popular network architectures and data sets, demonstrating the efficacy and accuracy of our coreset construction.

Skyline plays a pivotal role in mountainous visual geo-localization and localization/navigation of planetary rovers/UAVs and virtual/augmented reality applications. We present a novel mountainous skyline detection approach where we adapt a shallow learning approach to learn a set of filters to discriminate between edges belonging to sky-mountain boundary and others coming from different regions. Unlike earlier approaches, which either rely on extraction of explicit feature descriptors and their classification, or fine-tuning general scene parsing deep networks for sky segmentation, our approach learns linear filters based on local structure analysis. At test time, for every candidate edge pixel, a single filter is chosen from the set of learned filters based on pixel's structure tensor, and then applied to the patch around it. We then employ dynamic programming to solve the shortest path problem for the resultant multistage graph to get the sky-mountain boundary. The proposed approach is computationally faster than earlier methods while providing comparable performance and is more suitable for resource constrained platforms e.g., mobile devices, planetary rovers and UAVs. We compare our proposed approach against earlier skyline detection methods using four different data sets. Our code is available at \url{https://github.com/TouqeerAhmad/skyline_detection}.

Ross Blum, the CEO of Skyline Robotics a company that is automating robots in the workforce joins Enterprise Radio. Ross Blum joined Skyline as COO in December 2020. He will oversee day-to-day operations in Israel and the USA. Prior to joining Skyline, Ross served as the COO of Quidd, a a Sequoia Capital-backed startup. At Quidd, Ross developed and oversaw strategies across cross-functional teams to drive business results, managing all elements of day-to-day operations (Engineering, Product, Finance, HR, Content Production, and Partnerships).