Long-context inference is increasingly a memory-traffic problem. The culprit is the key--value (KV) cache: it grows with context length, batch size, layers, and heads, and it is read at every decoding step. Rotation-based scalar codecs meet this systems constraint by storing a norm, applying a shared random rotation, and quantizing one coordinate at a time. They are universal and random-access, but they discard the geometry created by the normalization step. After a Haar rotation, a block of $k$ consecutive coordinates is not a product source; it is a spherical-Beta source on the unit ball. We introduce \textsc{FibQuant}, a universal fixed-rate vector quantizer that keeps the same normalize--rotate--store interface while replacing scalar tables by a shared radial--angular codebook matched to this canonical source. The codebook combines Beta-quantile radii, Fibonacci\,/\,Roberts--Kronecker quasi-uniform directions, and multi-restart Lloyd--Max refinement. We prove that the resulting vector code strictly improves on its scalar product specialization at matched rate, with a high-rate gain that separates into a cell-shaping factor and a density-matching factor. The same construction gives a dense rate axis, including fractional-bit and sub-one-bit operating points, without calibration or variable-length addresses. On GPT-2 small KV caches, \textsc{FibQuant} traces a memory--fidelity frontier from $5\times$ compression at $0.99$ attention cosine similarity to $34\times$ at $0.95$. End-to-end on TinyLlama-1.1B, it is within $0.10$ perplexity of fp16 at $4\times$ compression and has $3.6\times$ lower perplexity than scalar \textsc{TurboQuant} at $b = 2$ ($8\times$ compression), where scalar random-access quantization begins to fail.

Activation functions play a central role in neural networks by shaping internal representations. Recently, learning binary activation representations has attracted significant attention due to their advantages in computational and memory efficiency, as well as interpretability. However, training neural networks with Heaviside activations remains challenging, as their non-differentiability obstructs standard gradient-based optimization. In this paper, we propose Heavy Tailed Activation Function (HTAF), a smooth approximation to the Heaviside function that enables stable training with gradient-based optimization. We construct HTAF as a sigmoid hyperbolic tangent composite function and theoretically show that it maintains a large gradient mass around zero inputs while exhibiting slower gradient decay in the tail regions. We show that Spiking Neural Networks, Binary Neural Networks and Deep Heaviside neural Networks can be trained stably using HTAF with gradient-based optimization. Finally, we introduce Implicit Concept Bottleneck Models (ICBMs), an interpretable image model that leverages HTAF to induce discrete feature representations. Extensive experiments across various architectures and image datasets demonstrate that ICBM enables stable discretization while achieving prediction performance comparable to or better than standard models.

Standard Bradley--Terry (BT) reward models are limited when human preferences are pluralistic. Although soft preference labels preserve disagreement information, BT can only express it by shrinking reward margins. Gaussian reward models provide an alternative by jointly predicting a reward mean and a reward variance, but suffer from a fundamental non-identifiability from pairwise preferences alone. We propose Anchor-guided Variance-aware Reward Modeling, a framework that resolves this non-identifiability by augmenting preference data with two coarse response-level anchor labels. Building on this, we prove that two anchors are sufficient for identification, develop a joint training objective and establish a non-asymptotic convergence rate for both the estimated reward mean and variance functions. Across simulation studies and four real-world diverging-preference datasets, our method consistently improves reward modeling performance and downstream RLHF, including PPO training and best-of-$N$ selection.

Uncertainty quantification has become an important factor in understanding the data representations produced by Graph Neural Networks (GNNs). Despite their predictive capabilities being ever useful across industrial workspaces, the inherent uncertainty induced by the nature of the data is a huge mitigating factor to GNN performance. While aleatoric uncertainty is the result of noisy and incomplete stochastic data such as missing edges or over-smoothing, epistemic uncertainty arises from lack of knowledge about a system or model (e.g., a graph's topology or node feature representation), which can be reduced by gathering more data and information. In this paper, we propose an original new framework in which node-level epistemic uncertainty is modelled in a belief function (finite random set) formalism. The resulting Random-Set Graph Neural Networks have a belief-function head predicting a random set over the list of classes, from which both a precise probability prediction and a measure of epistemic uncertainty can be obtained. Extensive experiments on 9 different graph learning datasets, including real-world autonomous driving benchmarks as such Nuscene and ROAD, demonstrate RS-GNN's superior uncertainty quantification capabilities.

We introduce Pion, a spectrum-preserving optimizer for large language model (LLM) training based on orthogonal equivalence transformation. Unlike additive optimizers such as Adam and Muon, Pion updates each weight matrix through left and right orthogonal transformations, preserving its singular values throughout training. This yields an optimization mechanism that modulates the geometry of weight matrices while keeping their spectral norm fixed. We derive the Pion update rule, systematically examine its design choices, and analyze its convergence behavior along with several key properties. Empirical results show that Pion offers a stable and competitive alternative to standard optimizers for both LLM pretraining and finetuning.

In a United States court, OpenAI chief executive Sam Altman has rejected claims from fellow tech mogul Elon Musk that he betrayed the artificial intelligence company's original vision. Tuesday marked the start of Altman's testimony in a contentious trial unfolding in Oakland, California, between some of tech's richest and most powerful titans. He alleged that OpenAI's leader persuaded him to invest $38bn, based on a goal of improving humanity, only to see the company pivot to a for-profit venture in 2019. On the witness stand on Tuesday, Altman instead framed Musk as a competitor obsessed with exercising control over OpenAI. "It does not fit with my conception of the words'stealing a charity' to look at what has actually happened here," Altman told the court.

Sam Altman is questioned by OpenAI's attorney, Bill Savitt, before Yvonne Gonzalez Rogers, a US district judge, at a federal courthouse in Oakland, California, on 12 May 2026 in a courtroom sketch. Sam Altman is questioned by OpenAI's attorney, Bill Savitt, before Yvonne Gonzalez Rogers, a US district judge, at a federal courthouse in Oakland, California, on 12 May 2026 in a courtroom sketch. The OpenAI CEO, Sam Altman, took the stand on Tuesday to defend himself and his company against a lawsuit by Elon Musk . Altman is set to be one of the final witnesses in the trial, which has pitted two of the tech industry's most powerful men against each other in a dramatic courtroom showdown. Musk has accused Altman and OpenAI of breaking the AI firm's founding agreement by restructuring it into a for-profit enterprise, alleging that Altman essentially swindled him into co-founding the company and providing tens of millions in financial backing.

Elon Musk tried to take control of OpenAI, even suggesting it could pass to his children when he dies, Sam Altman said on Tuesday. Altman is co-founder and chief executive of the artificial intelligence (AI) company behind ChatGPT. He is being sued by Musk, who accuses him of having looted a charity given OpenAI began as a non-profit. Appearing before a federal jury in Oakland, California, Altman said Musk not only backed the idea of OpenAI becoming a for-profit business, he wanted control of it for the long-run. A particularly hair-raising moment was when my cofounders asked, 'If you have control, what happens when you die?'

Google I/O, the company's big annual developer conference, is almost upon us . But the company isn't waiting until then to reveal what it has in store for Android. There was just far too much news on that front to squeeze into the I/O keynote, so Google revealed the details in the latest edition of The Android Show today. And, my goodness, were there a lot of details to reveal. From Gemini Intelligence and new laptops in the form of Googlebooks to an AirDrop-related update and Instagram editing tools in Android, Google had plenty of announcements to make. So, without further ado, here's an overview of everything Google announced during The Android Show: I/O edition.

You'll soon be able to generate your own widgets or ask Gemini to finish a booking in Chrome on Android. The Google I/O annual developer conference is around the corner--May 19--but in what is quickly becoming a tradition, Google announced new features for Android and Gemini a week early. The news came on Tuesday via the second-ever Android Show on YouTube . This livestreamed presentation helps Google spread out the cavalcade of updates from the often jam-packed I/O keynote. The Android Show focused on new features in Android 17, the next version of Android coming later this summer, as well as several updates to the Gemini assistant experience. It continues the theme set last year by Sameer Samat, president of the Android ecosystem, of turning Android into an "intelligent operating system."