In its early days, the AI initiative known as Project Maven had its fair share of skeptics at the Pentagon. Today, many of them are true believers. The rise of AI warfare speaks to the biggest moral and practical question there is: Who--or what--gets to decide to take a human life? And who bears that cost? In 2018, more than 3,000 Google workers protested the company's involvement in "the business of war" after finding out the company was part of Project Maven, then a nascent Pentagon effort to use computer vision to rifle through copious video footage taken in America's overseas drone wars. They feared Project Maven's AI could one day be used for lethal targeting. In my yearslong effort to uncover the full story of Project Maven for my book,, I learned that is exactly what happened, and that the undertaking was just as controversial inside the Pentagon. Today, the tool known as Maven Smart System is being used in US operations against Iran . How the US military's top brass moved from skepticism about the use of AI in war to true believers has a lot to do with a Marine colonel named Drew Cukor. In early September 2024, during the cocktail hour at a private retreat for tech investors and defense leaders, Vice Admiral Frank "Trey" Whitworth found his way to Drew Cukor. Now Project Maven's founding leader and his skeptical successor were standing face-to-face. Three years earlier, Whitworth had been the Pentagon's top military official for intelligence, advising the chairman of the Joint Chiefs of Staff and running one of the most sensitive and potentially lethal parts of any military process: targeting.

Though the US military's big data initiative Maven has sped up the planning of strikes for years, the comments suggest that generative AI is now adding a new interpretative layer to such deliberations. The US military might use generative AI systems to rank lists of targets and make recommendations--which would be vetted by humans--about which to strike first, according to a Defense Department official with knowledge of the matter. The disclosure about how the military may use AI chatbots comes as the Pentagon faces scrutiny over a strike on an Iranian school, which it is still investigating. A list of possible targets might be fed into a generative AI system that the Pentagon is fielding for classified settings. Then, said the official, who requested to speak on background with to discuss sensitive topics, humans might ask the system to analyze the information and prioritize the targets while accounting for factors like where aircraft are currently located. Humans would then be responsible for checking and evaluating the results and recommendations.

This dual approach bridges the semantic gap between visual and textual data, thereby improving the model's ability to process and interpret information from multiple images effectively.

Wemodel 34], whichisformallyG = hS, U, Pi. S is thestatespacet, every i 2 A {1,..., n} choosesui 2 U which action u 2 U Un. P(s0|s,u): S U S!

This paper presents MaVEn, an innovative Multi-granularity Visual Encoding framework designed to enhance the capabilities of Multimodal Large Language Models (MLLMs) in multi-image reasoning. Current MLLMs primarily focus on single-image visual understanding, limiting their ability to interpret and integrate information across multiple images.

Centralised training with decentralised execution is an important setting for cooperative deep multi-agent reinforcement learning due to communication constraints during execution and computational tractability in training. In this paper, we analyse value-based methods that are known to have superior performance in complex environments. We specifically focus on QMIX, the current state-of-the-art in this domain. We show that the representation constraints on the joint action-values introduced by QMIX and similar methods lead to provably poor exploration and suboptimality. Furthermore, we propose a novel approach called MAVEN that hybridises value and policy-based methods by introducing a latent space for hierarchical control. The value-based agents condition their behaviour on the shared latent variable controlled by a hierarchical policy. This allows MAVEN to achieve committed, temporally extended exploration, which is key to solving complex multi-agent tasks. Our experimental results show that MAVEN achieves significant performance improvements on the challenging SMAC domain.

Generalization across Agentic tool-calling environments remains a key unsolved challenge in developing reliable agentic reasoning systems. While large language models (LLMs) demonstrate strong performance on isolated benchmarks, their ability to transfer reasoning strategies and co-ordinate tools across diverse domains is poorly understood. In this work, we conduct a large-scale evaluation of state-of-the-art LLMs on multiple tool-calling benchmarksBFCL v3, TauBench, Tau2Bench, and AceBenchand introduce MAVEN (Math & Physics Adversarial Verification & Evaluation Network), a new out of distribution (OOD) benchmark designed to stress-test multi-step reasoning through explicit verification and adversarial task composition. Our results show that most current models achieve below 50% accuracy on MAVEN, revealing a significant generalization gap across tool-use settings. To address this, we present the CoreThink Agentic Reasoner, a framework that augments LLMs with a lightweight symbolic reasoning layer for structured decomposition and adaptive tool orchestration. Without additional training, it generalizes across all benchmarks, achieving state-of-the-art performance with 530% improvements over existing baselines at roughly one-tenth the computational cost.

This dual approach bridges the semantic gap between visual and textual data, thereby improving the model's ability to process and interpret information from multiple images effectively.

Neural Information Processing Systems http://nips.cc/