Human language and logic abilities are computationally quantified within the well-studied grammar-automata hierarchy. We identify three hierarchical tiers and two corresponding transitions and show their correspondence to specific abilities in transformer-based language models (LMs). These emergent abilities have often been described in terms of scaling; we show that it is the transition between tiers, rather than scaled size itself, that determines a system's capabilities. Specifically, humans effortlessly process language yet require critical training to perform arithmetic or logical reasoning tasks; and LMs possess language abilities absent from predecessor systems, yet still struggle with logical processing. We submit a novel benchmark of computational power, provide empirical evaluations of humans and fifteen LMs, and, most significantly, provide a theoretically grounded framework to promote careful thinking about these crucial topics. The resulting principled analyses provide explanatory accounts of the abilities and shortfalls of LMs, and suggest actionable insights into the expansion of their logic abilities.

RAG has proven its effectiveness in reducing hallucinations We go beyond accurate retrieval to emphasize in LLMs, when their knowledge is incomplete, robust generation that remains resilient to forgetting outdated, or lacks sufficient detail to accurately and distraction by the two challenges. Our key address specific queries (Gao et al., 2023b; idea for enhancing robustness is an extract-thengenerate Fan et al., 2024). A critical aspect of RAG is the approach, Ext2Gen, where the model "retrieval" process, which involves identifying and first extracts query-relevant sentences from the retrieved selecting relevant text chunks. The quality of these chunks and then refine the information to retrieved chunks plays a pivotal role in the overall generate a precise answer. The extraction step here performance of RAG, as they form the basis serves as a chain-of-thought (CoT) process (Wei for generating factual and contextually relevant answers et al., 2022; Chu et al., 2023), where the model provides aligned with the query intent (Asai et al., the evidence first before generating the final 2024; Wang et al., 2023; Zhang et al., 2024).

Large language models (LLMs) frequently hallucinate due to misaligned self-awareness, generating erroneous outputs when addressing queries beyond their knowledge boundaries. While existing approaches mitigate hallucinations via uncertainty estimation or query rejection, they suffer from computational inefficiency or sacrificed helpfulness. To address these issues, we propose the Explicit Knowledge Boundary Modeling (EKBM) framework, integrating fast and slow reasoning systems to harmonize reliability and usability. The framework first employs a fast-thinking model to generate confidence-labeled responses, enabling immediate use of high-confidence outputs. For uncertain predictions, a slow refinement model conducts targeted reasoning to improve accuracy. To align model behavior with our proposed object, we propose a hybrid training pipeline, enhancing self-awareness without degrading task performance. Evaluations on dialogue state tracking tasks demonstrate that EKBM achieves superior model reliability over uncertainty-based baselines. Further analysis reveals that refinement substantially boosts accuracy while maintaining low computational overhead. Our work establishes a scalable paradigm for advancing LLM reliability and balancing accuracy and practical utility in error-sensitive applications.

As the context window expands, self-attention increasingly dominates the transformer's inference time. Therefore, accelerating attention computation while minimizing performance degradation is essential for the efficient deployment of Large Language Models (LLMs). In this study we extend a theoretical framework of attention sparsity in LLMs. A customized loss function is designed to enforce the sparsity by restricting the number of top elements in the attention matrix. We perform an initial set of evaluations with GPT-2 to show the effectiveness of our sparsification approach. The attention matrices of the models trained with the proposed loss are both sparse and effective in capturing relevant input dependencies. We now continue working to demonstrate the value of our approach on larger models and different architectures.

We address the neural network robustness problem by adding Similarity (i.e., correctly predicted depth-matches into training)-awareness and Distance-to-training-distribution-awareness to the existing output Magnitude (i.e., decision-boundary)-awareness of the softmax function. The resulting sdm activation function provides strong signals of the relative epistemic (reducible) predictive uncertainty. We use this novel behavior to further address the complementary HCI problem of mapping the output to human-interpretable summary statistics over relevant partitions of a held-out calibration set. Estimates of prediction-conditional uncertainty are obtained via a parsimonious learned transform over the class-conditional empirical CDFs of the output of a final-layer sdm activation function. For decision-making and as an intrinsic model check, estimates of class-conditional accuracy are obtained by further partitioning the high-probability regions of this calibrated output into class-conditional, region-specific CDFs. The uncertainty estimates from sdm calibration are remarkably robust to test-time distribution shifts and out-of-distribution inputs; incorporate awareness of the effective sample size; provide estimates of uncertainty from the learning and data splitting processes; and are well-suited for selective classification and conditional branching for additional test-time compute based on the predictive uncertainty, as for selective LLM generation, routing, and composition over multiple models and retrieval. Finally, we construct sdm networks, LLMs with uncertainty-aware verification and interpretability-by-exemplar as intrinsic properties. We provide open-source software implementing these results.

Chain-of-Thought (CoT) significantly enhances the performance of large language models (LLMs) across a wide range of tasks, and prior research shows that CoT can theoretically increase expressiveness. However, there is limited mechanistic understanding of the algorithms that Transformer+CoT can learn. In this work, we (1) evaluate the state tracking capabilities of Transformer+CoT and its variants, confirming the effectiveness of CoT. (2) Next, we identify the circuit, a subset of model components, responsible for tracking the world state, finding that late-layer MLP neurons play a key role. We propose two metrics, compression and distinction, and show that the neuron sets for each state achieve nearly 100% accuracy, providing evidence of an implicit finite state automaton (FSA) embedded within the model. (3) Additionally, we explore three realistic settings: skipping intermediate steps, introducing data noise, and testing length generalization. Our results demonstrate that Transformer+CoT learns robust algorithms (FSA), highlighting its resilience in challenging scenarios.

Multiple-choice benchmarks, consisting of various prompts and choices, are among the most widely used methods to assess a language model's natural language understanding capability. Given a specific prompt, we typically compute $P(Choice|Prompt)$ to evaluate how likely a language model is to generate the correct choice compared to incorrect ones. However, we observe that performance measured using this approach reflects not only the model's comprehension of the prompt but also its inherent biases for certain choices regardless of the prompt. This issue makes it challenging to accurately measure a model's natural language understanding, as models may select the answer without fully understanding the prompt. To address this limitation, we propose a novel metric called ANPMI, which normalizes Pointwise Mutual Information (PMI) by $-\log P(Choice)$. ANPMI provides a more accurate assessment of the model's natural language understanding by ensuring that it is challenging to answer a question without properly understanding the prompt.

Flow matching has emerged as a powerful framework for generative modeling, offering computational advantages over diffusion models by leveraging deterministic Ordinary Differential Equations (ODEs) instead of stochastic dynamics. While prior work established the worst case optimality of standard flow matching under Wasserstein distances, the theoretical guarantees for higher-order flow matching - which incorporates acceleration terms to refine sample trajectories - remain unexplored. In this paper, we bridge this gap by proving that higher-order flow matching preserves worst case optimality as a distribution estimator. We derive upper bounds on the estimation error for second-order flow matching, demonstrating that the convergence rates depend polynomially on the smoothness of the target distribution (quantified via Besov spaces) and key parameters of the ODE dynamics. Our analysis employs neural network approximations with carefully controlled depth, width, and sparsity to bound acceleration errors across both small and large time intervals, ultimately unifying these results into a general worst case optimal bound for all time steps.

First, let's get the pesky business of defining AGI out of the way. In practice, it's a deeply hazy and changeable term shaped by the researchers or companies set on building the technology. But it usually refers to a future AI that outperforms humans on cognitive tasks. Which humans and which tasks we're talking about makes all the difference in assessing AGI's achievability, safety, and impact on labor markets, war, and society. That's why defining AGI, though an unglamorous pursuit, is not pedantic but actually quite important, as illustrated in a new paper published this week by authors from Hugging Face and Google, among others.

For some years now, China's annual gathering of its national legislature had been an increasingly disciplined and choreographed affair -- its muted vibes practically an echo of deepening concern about domestic stagnation. The National People's Congress seven-day gathering, which concludes Tuesday in Beijing, came on the heels of a breakthrough in artificial intelligence by China's home-grown startup DeepSeek that's fired up investors, politicians and even regulators. It also followed Chinese President Xi Jinping's high-profile meeting with business chiefs including Jack Ma.