An AI model released by the Chinese AI company DeepSeek uses new techniques that could significantly improve AI's ability to "remember." Released last week, the optical character recognition (OCR) model works by extracting text from an image and turning it into machine-readable words. This is the same technology that powers scanner apps, translation of text in photos, and many accessibility tools. OCR is already a mature field with numerous high-performing systems, and according to the paper and some early reviews, DeepSeek's new model performs on par with top models on key benchmarks. But researchers say the model's main innovation lies in how it processes information--specifically, how it stores and retrieves memories. Improving how AI models "remember" information could reduce the computing power they need to run, thus mitigating AI's large (and growing) carbon footprint.

The AI job cuts are here - or are they? Amazon's move this week to slash thousands of corporate jobs fed into a longstanding anxiety: that Artificial Intelligence is starting to replace workers. The tech giant joined a growing list of companies in the US that have pointed to AI technology as a reason behind layoffs. But some question whether AI is fully to blame - and have voiced scepticism that recent high-profile layoffs are a telling sign of the technology's effect on employment. Chegg, the online education firm, cited the new realities of AI as it announced a 45% reduction in workforce on Monday.

Modern LLMs are increasingly deep, and depth correlates with performance, albeit with diminishing returns. However, do these models use their depth efficiently? Do they compose more features to create higher-order computations that are impossible in shallow models, or do they merely spread the same kinds of computation out over more layers? To address these questions, we analyze the residual stream of the Llama 3.1, Qwen 3, and OLMo 2 family of models. We find: First, comparing the output of the sublayers to the residual stream reveals that layers in the second half contribute much less than those in the first half, with a clear phase transition between the two halves. Second, skipping layers in the second half has a much smaller effect on future computations and output predictions. Third, for multihop tasks, we are unable to find evidence that models are using increased depth to compose subresults in examples involving many hops. Fourth, we seek to directly address whether deeper models are using their additional layers to perform new kinds of computation. To do this, we train linear maps from the residual stream of a shallow model to a deeper one. We find that layers with the same relative depth map best to each other, suggesting that the larger model simply spreads the same computations out over its many layers. All this evidence suggests that deeper models are not using their depth to learn new kinds of computation, but only using the greater depth to perform more fine-grained adjustments to the residual. This may help explain why increasing scale leads to diminishing returns for stacked Transformer architectures.

Emerging large language model (LLM) applications involve diverse reasoning strategies and agentic workflows, straining the capabilities of existing serving systems built on a monolithic token generation loop. This paper introduces Pie, a programmable LLM serving system designed for flexibility and efficiency. Pie decomposes the traditional generation loop into fine-grained service handlers exposed via an API and delegates control of the generation process to user-provided programs, called inferlets. This enables applications to implement new KV cache strategies, bespoke generation logic, and seamlessly integrate computation and I/O-entirely within the application, without requiring modifications to the serving system. Pie executes inferlets using WebAssembly, benefiting from its lightweight sandboxing. Our evaluation shows Pie matches state-of-the-art performance on standard tasks (3-12% latency overhead) while significantly improving latency and throughput (1.3x-3.4x higher) on agentic workflows by enabling application-specific optimizations.

Deep neural networks (DNNs) have been used to model complex optimization problems in many applications, yet have difficulty guaranteeing solution optimality and feasibility, despite training on large datasets. Training a NN as a surrogate optimization solver amounts to estimating a global solution function that maps varying problem input parameters to the corresponding optimal solutions. Work in multiparametric programming (mp) has shown that solutions to quadratic programs (QP) are piece-wise linear functions of the parameters, and researchers have suggested leveraging this property to model mp-QP using NN with ReLU activation functions, which also exhibit piecewise linear behaviour. This paper proposes a NN modeling approach and learning algorithm that discovers the exact closed-form solution to QP with linear constraints, by analytically deriving NN model parameters directly from the problem coefficients without training. Whereas generic DNN cannot guarantee accuracy outside the training distribution, the closed-form NN model produces exact solutions for every discovered critical region of the solution function. To evaluate the closed-form NN model, it was applied to DC optimal power flow problems in electricity management. In terms of Karush-Kuhn-Tucker (KKT) optimality and feasibility of solutions, it outperformed a classically trained DNN and was competitive with, or outperformed, a commercial analytic solver (Gurobi) at far less computational cost. For a long-range energy planning problem, it was able to produce optimal and feasible solutions for millions of input parameters within seconds.

Variational mean field approximations tend to struggle with contemporary overparametrized deep neural networks. Where a Bayesian treatment is usually associated with high-quality predictions and uncertainties, the practical reality has been the opposite, with unstable training, poor predictive power, and subpar calibration. Building upon recent work on reparametrizations of neural networks, we propose a simple variational family that considers two independent linear subspaces of the parameter space. These represent functional changes inside and outside the support of training data. This allows us to build a fully-correlated approximate posterior reflecting the overparametrization that tunes easy-to-interpret hyperparameters. We develop scalable numerical routines that maximize the associated evidence lower bound (ELBO) and sample from the approximate posterior. Empirically, we observe state-of-the-art performance across tasks, models, and datasets compared to a wide array of baseline methods. Our results show that approximate Bayesian inference applied to deep neural networks is far from a lost cause when constructing inference mechanisms that reflect the geometry of reparametrizations.

Reinforcement Learning from Human Feedback (RLHF) has emerged as a key technique for post-training large language models. Despite its empirical success, the theoretical understanding of RLHF is still limited, as learning the KL-regularized target with only preference feedback poses additional challenges compared with canonical RL. Existing works mostly study the reward-based Bradley-Terry (BT) preference model, and extend classical designs utilizing optimism or pessimism. This work, instead, considers the general preference model (whose practical relevance has been observed recently) and obtains performance guarantees with major, order-wise improvements over existing ones. Surprisingly, these results are derived from algorithms that directly use the empirical estimates (i.e., greedy sampling), as opposed to constructing optimistic or pessimistic estimates in previous works. This insight has a deep root in the unique structural property of the optimal policy class under the KL-regularized target, and we further specialize it to the BT model, highlighting the surprising sufficiency of greedy sampling in RLHF.

Processing structured tabular data, particularly large and lengthy tables, constitutes a fundamental yet challenging task for large language models (LLMs). However, existing long-context benchmarks like Needle-in-a-Haystack primarily focus on unstructured text, neglecting the challenge of diverse structured tables. Meanwhile, previous tabular benchmarks mainly consider downstream tasks that require high-level reasoning abilities, and overlook models' underlying fine-grained perception of individual table cells, which is crucial for practical and robust LLM-based table applications. To address this gap, we introduce \textsc{NeedleInATable} (NIAT), a new long-context tabular benchmark that treats each table cell as a ``needle'' and requires models to extract the target cell based on cell locations or lookup questions. Our comprehensive evaluation of various LLMs and multimodal LLMs reveals a substantial performance gap between popular downstream tabular tasks and the simpler NIAT task, suggesting that they may rely on dataset-specific correlations or shortcuts to obtain better benchmark results but lack truly robust long-context understanding towards structured tables. Furthermore, we demonstrate that using synthesized NIAT training data can effectively improve performance on both NIAT task and downstream tabular tasks, which validates the importance of NIAT capability for LLMs' genuine table understanding ability.

In-context reinforcement learning (ICRL) has emerged as a promising paradigm for adapting RL agents to downstream tasks through prompt conditioning. However, two notable challenges remain in fully harnessing in-context learning within RL domains: the intrinsic multi-modality of the state-action-reward data and the diverse, heterogeneous nature of decision tasks. To tackle these challenges, we propose T2MIR (Token- and Task-wise MoE for In-context RL), an innovative framework that introduces architectural advances of mixture-of-experts (MoE) into transformer-based decision models. T2MIR substitutes the feedforward layer with two parallel layers: a token-wise MoE that captures distinct semantics of input tokens across multiple modalities, and a task-wise MoE that routes diverse tasks to specialized experts for managing a broad task distribution with alleviated gradient conflicts. To enhance task-wise routing, we introduce a contrastive learning method that maximizes the mutual information between the task and its router representation, enabling more precise capture of task-relevant information. The outputs of two MoE components are concatenated and fed into the next layer. Comprehensive experiments show that T2MIR significantly facilitates in-context learning capacity and outperforms various types of baselines. We bring the potential and promise of MoE to ICRL, offering a simple and scalable architectural enhancement to advance ICRL one step closer toward achievements in language and vision communities. Our code is available at https://github.com/NJU-RL/T2MIR.

Accurate prediction of ionic conductivity in electrolyte systems is crucial for advancing numerous scientific and technological applications. While significant progress has been made, current research faces two fundamental challenges: (1) the lack of high-quality standardized benchmarks, and (2) inadequate modeling of geometric structure and intermolecular interactions in mixture systems. To address these limitations, we first reorganize and enhance the CALiSol and DiffMix electrolyte datasets by incorporating geometric graph representations of molecules. We then propose GeoMix, a novel geometry-aware framework that preserves Set-SE(3) equivariance-an essential but challenging property for mixture systems. At the heart of GeoMix lies the Geometric Interaction Network (GIN), an equivariant module specifically designed for intermolecular geometric message passing. Comprehensive experiments demonstrate that GeoMix consistently outperforms diverse baselines (including MLPs, GNNs, and geometric GNNs) across both datasets, validating the importance of cross-molecular geometric interactions and equivariant message passing for accurate property prediction. This work not only establishes new benchmarks for electrolyte research but also provides a general geometric learning framework that advances modeling of mixture systems in energy materials, pharmaceutical development, and beyond.