Against those expectations, the model has mostly underwhelmed. People have highlighted glaring mistakes in GPT-5's responses, countering Altman's claim made at the launch that it works like "a legitimate PhD-level expert in anything any area you need on demand." Early testers have also found issues with OpenAI's promise that GPT-5 automatically works out what type of AI model is best suited for your question--a reasoning model for more complicated queries, or a faster model for simpler ones. Altman seems to have conceded that this feature is flawed and takes away user control. However there is good news too: the model seems to have eased the problem of ChatGPT sucking up to users, with GPT-5 less likely to shower them with over the top compliments.

As large language models continue to scale up in both size and context length, the memory and communication cost of key-value (KV) cache storage has become a major bottleneck in multi-GPU and multi-node inference. While MoE-based architectures sparsify computation across experts, the corresponding KV caches remain dense and globally synchronized, resulting in significant overhead. We introduce \textbf{PiKV}, a parallel and distributed KV cache serving framework tailored for MoE architecture. PiKV leverages \textit{expert-sharded KV storage} to partition caches across GPUs, \textit{PiKV routing} to reduce token-to-KV access, and a \textit{PiKV Scheduling} to adaptively retain query-relevant entries. To further reduce memory usage, PiKV integrates \textit{PiKV Compression} modules the caching pipeline for acceleration. PiKV is recently publicly available as an open-source software library: \href{https://github.com/NoakLiu/PiKV}{https://github.com/NoakLiu/PiKV}. Experiments details is recorded at: \href{https://github.com/NoakLiu/PiKV/blob/main/downstream_tasks/README.md}{https://github.com/NoakLiu/PiKV/Experimental\_Results}. We also have PiKV integrated with Nvidia kvpress for acceleration, details see \href{https://github.com/NoakLiu/PiKVpress}{https://github.com/NoakLiu/PiKVpress}. PiKV is still a living project, aiming to become a comprehesive KV Cache management system for MoE Architectures.

Recent advances in learning rate (LR) scheduling have demonstrated the effectiveness of decay-free approaches that eliminate the traditional decay phase while maintaining competitive performance. Model merging techniques have emerged as particularly promising solutions in this domain. We present Warmup-Stable and Merge (WSM), a general framework that establishes a formal connection between learning rate decay and model merging. WSM provides a unified theoretical foundation for emulating various decay strategies-including cosine decay, linear decay and inverse square root decay-as principled model averaging schemes, while remaining fully compatible with diverse optimization methods. Through extensive experiments, we identify merge duration-the training window for checkpoint aggregation-as the most critical factor influencing model performance, surpassing the importance of both checkpoint interval and merge quantity. Our framework consistently outperforms the widely-adopted Warmup-Stable-Decay (WSD) approach across multiple benchmarks, achieving significant improvements of +3.5% on MATH, +2.9% on HumanEval, and +5.5% on MMLU-Pro. The performance advantages extend to supervised fine-tuning scenarios, highlighting WSM's potential for long-term model refinement.

Predictions and generations from large language models are increasingly being explored as an aid to computational social science and human subject research in limited data regimes. While previous technical work has explored the potential to use model-predicted labels for unlabeled data in a principled manner, there is increasing interest in using large language models to generate entirely new synthetic samples (also termed as synthetic simulations), such as in responses to surveys. However, it is not immediately clear by what means practitioners can combine such data with real data and yet produce statistically valid conclusions upon them. In this work, we introduce a new estimator based on generalized method of moments, providing a hyperparameter-free solution with strong theoretical guarantees to address the challenge at hand. Surprisingly, we find that interactions between the moment residuals of synthetic data and those of real data can improve estimates of the target parameter. We empirically validate the finite-sample performance of our estimator across different regression tasks in computational social science applications, demonstrating large empirical gains.

We present Zero-Direction Probing (ZDP), a theory-only framework for detecting model drift from null directions of transformer activations without task labels or output evaluations. Under assumptions A1--A6, we prove: (i) the Variance--Leak Theorem, (ii) Fisher Null-Conservation, (iii) a Rank--Leak bound for low-rank updates, and (iv) a logarithmic-regret guarantee for online null-space trackers. We derive a Spectral Null-Leakage (SNL) metric with non-asymptotic tail bounds and a concentration inequality, yielding a-priori thresholds for drift under a Gaussian null model. These results show that monitoring right/left null spaces of layer activations and their Fisher geometry provides concrete, testable guarantees on representational change.

Large language models (LLMs) generate text embeddings from text data, producing vector representations that capture the semantic meaning and contextual relationships of words. However, the high dimensionality of these embeddings often impedes efficiency and drives up computational cost in downstream tasks. To address this, we propose AutoEncoder-Augmented Learning with Text (AEALT), a supervised, factor-augmented framework that incorporates dimension reduction directly into pre-trained LLM workflows. First, we extract embeddings from text documents; next, we pass them through a supervised augmented autoencoder to learn low-dimensional, task-relevant latent factors. By modeling the nonlinear structure of complex embeddings, AEALT outperforms conventional deep-learning approaches that rely on raw embeddings. We validate its broad applicability with extensive experiments on classification, anomaly detection, and prediction tasks using multiple real-world public datasets. Numerical results demonstrate that AEALT yields substantial gains over both vanilla embeddings and several standard dimension reduction methods.

Congenital heart disease (CHD) presents complex, lifelong challenges often underrepresented in traditional clinical metrics. While unstructured narratives offer rich insights into patient and caregiver experiences, manual thematic analysis (TA) remains labor-intensive and unscalable. We propose a fully automated large language model (LLM) pipeline that performs end-to-end TA on clinical narratives, which eliminates the need for manual coding or full transcript review. Our system employs a novel multi-agent framework, where specialized LLM agents assume roles to enhance theme quality and alignment with human analysis. To further improve thematic relevance, we optionally integrate reinforcement learning from human feedback (RLHF). This supports scalable, patient-centered analysis of large qualitative datasets and allows LLMs to be fine-tuned for specific clinical contexts.

Participatory Budgeting (PB) empowers citizens to propose and vote on public investment projects. Yet, despite its democratic potential, PB initiatives often suffer from low participation rates, limiting their visibility and perceived legitimacy. In this work, we aim to strengthen PB elections in two key ways: by supporting project proposers in crafting better proposals, and by helping PB organizers manage large volumes of submissions in a transparent manner. We propose a privacy-preserving approach to predict which PB proposals are likely to be funded, using only their textual descriptions and anonymous historical voting records -- without relying on voter demographics or personally identifiable information. We evaluate the performance of GPT 4 Turbo in forecasting proposal outcomes across varying contextual scenarios, observing that the LLM's prior knowledge needs to be complemented by past voting data to obtain predictions reflecting real-world PB voting behavior. Our findings highlight the potential of AI-driven tools to support PB processes by improving transparency, planning efficiency, and civic engagement.

Large Language Models (LLMs) struggle with dynamic knowledge updates, especially when new information conflicts with deeply embedded facts. Such conflicting factual edits often lead to two critical issues: resistance to adopting the new fact and severe catastrophic forgetting of unrelated knowledge. This paper introduces and evaluates a novel "unlearn-then-learn" strategy for precise knowledge editing in LLMs, leveraging the parameter-efficient fine-tuning (PEFT) technique, Infused Adapter by Inhibiting and Amplifying Inner Activations ($IA^3$). Crucially, this two-stage approach is powered by an initial circuit localization phase that identifies and targets the specific internal components responsible for encoding the conflicting fact. Through a rigorous experimental methodology on microsoft/Phi-3-mini-4k-instruct, we demonstrate that this mechanistically informed two-stage approach achieves near-perfect accuracy (98.50%) for the new, modulated fact while simultaneously effectively suppressing the original conflicting fact (96.00% forget rate). Critically, our strategy exhibits unprecedented localization (72.00% F_control accuracy), dramatically mitigating catastrophic forgetting observed in direct fine-tuning approaches (which showed as low as ~20% F_control accuracy), a direct benefit of our targeted interpretability-guided intervention. Furthermore, qualitative analysis reveals a nuanced mechanism of "soft forgetting," where original knowledge is suppressed from default retrieval but remains latent and conditionally accessible, enhancing model safety and control. These findings represent a significant advancement towards precise, localized, and safe knowledge management in compact LLMs.

With the breakthrough progress of large language models (LLMs) in natural language processing and multimodal tasks, efficiently deploying them on resource-constrained edge devices has become a critical challenge. The Mixture of Experts (MoE) architecture enhances model capacity through sparse activation, but faces two major difficulties in practical deployment: (1) The presence of numerous outliers in activation distributions leads to severe degradation in quantization accuracy for both activations and weights, significantly impairing inference performance; (2) Under limited memory, efficient offloading and collaborative inference of expert modules struggle to balance latency and throughput. To address these issues, this paper proposes an efficient MoE edge deployment scheme based on Hessian-Aware Quantization (HAQ) and CPU-GPU collaborative inference. First, by introducing smoothed Hessian matrix quantization, we achieve joint 8-bit quantization of activations and weights, which significantly alleviates the accuracy loss caused by outliers while ensuring efficient implementation on mainstream hardware. Second, we design an expert-level collaborative offloading and inference mechanism, which, combined with expert activation path statistics, enables efficient deployment and scheduling of expert modules between CPU and GPU, greatly reducing memory footprint and inference latency. Extensive experiments validate the effectiveness of our method on mainstream large models such as the OPT series and Mixtral 8*7B: on datasets like Wikitext2 and C4, the inference accuracy of the low-bit quantized model approaches that of the full-precision model, while GPU memory usage is reduced by about 60%, and inference latency is significantly improved.