When Reward Hacking Rebounds: Understanding and Mitigating It with Representation-Level Signals
arXiv:2604.01476v1 Announce Type: new Abstract: Reinforcement learning for LLMs is vulnerable to reward hacking, where models exploit shortcuts to maximize reward without solving the intended task. We systematically study this phenomenon in coding tasks using an environment-manipulation setting, where models...
Efficient and Principled Scientific Discovery through Bayesian Optimization: A Tutorial
arXiv:2604.01328v1 Announce Type: new Abstract: Traditional scientific discovery relies on an iterative hypothesise-experiment-refine cycle that has driven progress for centuries, but its intuitive, ad-hoc implementation often wastes resources, yields inefficient designs, and misses critical insights. This tutorial presents Bayesian Optimisation...
Asymmetric Actor-Critic for Multi-turn LLM Agents
arXiv:2604.00304v1 Announce Type: new Abstract: Large language models (LLMs) exhibit strong reasoning and conversational abilities, but ensuring reliable behavior in multi-turn interactions remains challenging. In many real-world applications, agents must succeed in one-shot settings where retries are impossible. Existing approaches...
OmniVoice: Towards Omnilingual Zero-Shot Text-to-Speech with Diffusion Language Models
arXiv:2604.00688v2 Announce Type: new Abstract: We present OmniVoice, a massive multilingual zero-shot text-to-speech (TTS) model that scales to over 600 languages. At its core is a novel diffusion language model-style discrete non-autoregressive (NAR) architecture. Unlike conventional discrete NAR models that...
The Chronicles of RiDiC: Generating Datasets with Controlled Popularity Distribution for Long-form Factuality Evaluation
arXiv:2604.00019v1 Announce Type: cross Abstract: We present a configurable pipeline for generating multilingual sets of entities with specified characteristics, such as domain, geographical location and popularity, using data from Wikipedia and Wikidata. These datasets are intended for evaluating the factuality...
Logarithmic Scores, Power-Law Discoveries: Disentangling Measurement from Coverage in Agent-Based Evaluation
arXiv:2604.00477v1 Announce Type: new Abstract: LLM-based agent judges are an emerging approach to evaluating conversational AI, yet a fundamental uncertainty remains: can we trust their assessments, and if so, how many are needed? Through 960 sessions with two model pairs...
Benchmark for Assessing Olfactory Perception of Large Language Models
arXiv:2604.00002v1 Announce Type: cross Abstract: Here we introduce the Olfactory Perception (OP) benchmark, designed to assess the capability of large language models (LLMs) to reason about smell. The benchmark contains 1,010 questions across eight task categories spanning odor classification, odor...
Variational LSTM with Augmented Inputs: Nonlinear Response History Metamodeling with Aleatoric and Epistemic Uncertainty
arXiv:2604.01587v1 Announce Type: new Abstract: Uncertainty propagation in high-dimensional nonlinear dynamic structural systems is pivotal in state-of-the-art performance-based design and risk assessment, where uncertainties from both excitations and structures, i.e., the aleatoric uncertainty, must be considered. This poses a significant...
Dynin-Omni: Omnimodal Unified Large Diffusion Language Model
arXiv:2604.00007v1 Announce Type: cross Abstract: We present Dynin-Omni, the first masked-diffusion-based omnimodal foundation model that unifies text, image, and speech understanding and generation, together with video understanding, within a single architecture. Unlike autoregressive unified models that serialize heterogeneous modalities, or...
MSA-Thinker: Discrimination-Calibration Reasoning with Hint-Guided Reinforcement Learning for Multimodal Sentiment Analysis
arXiv:2604.00013v1 Announce Type: cross Abstract: Multimodal sentiment analysis aims to understand human emotions by integrating textual, auditory, and visual modalities. Although Multimodal Large Language Models (MLLMs) have achieved state-of-the-art performance via supervised fine-tuning (SFT), their end-to-end "black-box" nature limits interpretability....
Brevity Constraints Reverse Performance Hierarchies in Language Models
arXiv:2604.00025v1 Announce Type: new Abstract: Standard evaluation protocols reveal a counterintuitive phenomenon: on 7.7% of benchmark problems spanning five datasets, larger language models underperform smaller ones by 28.4 percentage points despite 10-100x more parameters. Through systematic evaluation of 31 models...
Therefore I am. I Think
arXiv:2604.01202v2 Announce Type: new Abstract: We consider the question: when a large language reasoning model makes a choice, did it think first and then decide to, or decide first and then think? In this paper, we present evidence that detectable,...
Microsoft takes on AI rivals with three new foundational models
MAI released models that can transcribe voice into text as well as generate audio and images after the group's formation six months ago.
A Taxonomy of Programming Languages for Code Generation
arXiv:2604.00239v1 Announce Type: new Abstract: The world's 7,000+ languages vary widely in the availability of resources for NLP, motivating efforts to systematically categorize them by their degree of resourcefulness (Joshi et al., 2020). A similar disparity exists among programming languages...
Criterion Validity of LLM-as-Judge for Business Outcomes in Conversational Commerce
arXiv:2604.00022v1 Announce Type: cross Abstract: Multi-dimensional rubric-based dialogue evaluation is widely used to assess conversational AI, yet its criterion validity -- whether quality scores are associated with the downstream outcomes they are meant to serve -- remains largely untested. We...
OpenAI, not yet public, raises $3B from retail investors in monster $122B fund raise
OpenAI's latest funding round, led by Amazon, Nvidia, and SoftBank, values the AI lab at $852 billion as it nears an IPO.
Locally Confident, Globally Stuck: The Quality-Exploration Dilemma in Diffusion Language Models
arXiv:2604.00375v1 Announce Type: new Abstract: Diffusion large language models (dLLMs) theoretically permit token decoding in arbitrary order, a flexibility that could enable richer exploration of reasoning paths than autoregressive (AR) LLMs. In practice, however, random-order decoding often hurts generation quality....
Do Language Models Know When They'll Refuse? Probing Introspective Awareness of Safety Boundaries
arXiv:2604.00228v1 Announce Type: new Abstract: Large language models are trained to refuse harmful requests, but can they accurately predict when they will refuse before responding? We investigate this question through a systematic study where models first predict their refusal behavior,...
Malliavin Calculus for Counterfactual Gradient Estimation in Adaptive Inverse Reinforcement Learning
arXiv:2604.01345v1 Announce Type: new Abstract: Inverse reinforcement learning (IRL) recovers the loss function of a forward learner from its observed responses adaptive IRL aims to reconstruct the loss function of a forward learner by passively observing its gradients as it...
Cognichip wants AI to design the chips that power AI, and just raised $60M to try
The firm says it can reduce the cost of chip development by more than 75% and cut the timeline by more than half.
AI Company Safety Practices Fall Short of Public Commitments and Show Structural Weaknesses, as Top Performers Widen the Gap
But in a win for transparency, five leading companies participated in the scorecard's survey for the first time, providing critical new information to the public.
Judge irate as defendant joins by Zoom while driving—then lies about it
"Let me see the driver!"
AV1’s open, royalty-free promise in question as Dolby sues Snapchat over codec
Big Tech declaring AV1 royalty-free “doesn't mean that it is."
Senators want US energy information agency to monitor data center electricity usage
In a letter, senators press for mandated annual electricity disclosure for data centers.
Anthropic’s Claude popularity with paying consumers is skyrocketing
Estimates for total Claude consumer users are all over the map (we've seen figures ranging from 18 million to 30 million). Anthropic hasn't disclosed this data, but a spokesperson did tell TechCrunch that Claude paid subscriptions have more than doubled...
VCs are betting billions on AI’s next wave, so why is OpenAI killing Sora?
When an 82-year-old Kentucky woman was offered $26 million from an AI company that wanted to build a data center on her land, she said no. Sure, that same company can try to rezone 2,000 acres nearby anyway, but as...
Wikipedia cracks down on the use of AI in article writing
The site, whose policies are subject to change, has struggled with the issue of AI-generated writing.
Cohere launches an open source voice model specifically for transcription
Relatively light at just 2 billion parameters, the model is meant for use with consumer-grade GPUs for those who want to self-host it. It currently supports 14 languages.
Corrigendum to “Generating dynamic lip-syncing using target audio in a multimedia environment” [Natural Language Processing Journal, Volume 8, 2024]
Do 3D Large Language Models Really Understand 3D Spatial Relationships?
arXiv:2603.23523v1 Announce Type: new Abstract: Recent 3D Large-Language Models (3D-LLMs) claim to understand 3D worlds, especially spatial relationships among objects. Yet, we find that simply fine-tuning a language model on text-only question-answer pairs can perform comparably or even surpass these...