News

A new arXiv preprint (2608.00301v1) proves that error-penalized scoring rules, which reward correct answers with +1, penalize wrong ones with -λ, and give 0 for abstention, can cause KL-anchored gradient learners to drif…

Researchers introduced a flow-map learning (FML) framework to model unknown nonlocal partial differential equations (PDEs) directly from solution data, bypassing explicit evaluation of nonlocal operators. The method lear…

Researchers introduced DLLM-TTS, a text-to-speech framework using block discrete diffusion over X-Codec2 neural audio codec tokens, achieving a real-time factor of 0.15 with a 0.6B-parameter model trained on 20K hours, a…

Researchers introduced LLantia, an automated method using large language models to infer neural circuit function from connectome data, and applied it to the adult fruit fly brain, inferring cell type functions for all ce…

Researchers at arXiv introduced a novel pre-training method for logographic character recognition that uses semantic-based contrastive learning to improve visual representations on imbalanced datasets. The method, which …

Researchers introduced DE-NER, a dialogue elicitation framework that improves zero-shot Named Entity Recognition by leveraging the chatting ability of Large Language Models, achieving an average improvement of 3.75% F1 p…

Researchers propose a multimodal machine-learning framework for calibration monitoring and quality assessment in archaeological digitisation, integrating photogrammetric 3D reconstruction, hyperspectral imaging, X-ray fl…

A new arXiv study (2608.00077v1) audits visual-token pruning in OCR-critical multimodal large language models (MLLMs), finding that accuracy alone misses failures where correct answers lack local token provenance. On loc…

Researchers propose DAR, a reference-guided renderer that extends Wan2.2 camera control to a joint camera-plus-geometry interface, enabling pretrained video diffusion models to act as 4D renderers from animated meshes. O…

A new paper on arXiv introduces Context Compaction Theory, the first formal framework for analyzing how AI agents compress their state to fit within LLM context windows. The authors prove an equivalence between the Conte…

MoonPay launched PayBox on July 23, going live July 28, to let ChatGPT and Claude execute crypto and card payments using passkey authentication and user-set spending limits, with Fortune reporting the tool aims to make A…

Kevin Rose released Reflection Engine v1.3, a free prompt that asks an AI assistant to mine its memory of a user for blind spots, contradictions, emotional habits and likely trajectories, distributed via GitHub and X on …

Fifteen state attorneys general have sent a letter to OpenAI, signaling a coordinated regulatory push focused on consumer protection, data privacy, and potential harms from deployed AI models, as federal rules remain slo…

A developer's benchmark of LLM inference on consumer hardware reveals that prefill speed, not generation, is the bottleneck for large prompts, with an 18-fold variation across machines. The test also shows that AMD's 'AI…

A developer using OpenAI's GPT-5 (referred to as 'Sol') generated a working Three.js Matrix rain effect in a single prompt, producing about 300 lines of self-contained HTML code that ran on the first try. The developer r…

Graduate job vacancies in the UK have fallen to their lowest level since 2020, with Indeed reporting a 7% year-on-year decline, while AI skills demand reached a record high, appearing in 9.4% of UK job ads by end of June…

A developer credits Claude with improving architectural decisions before coding, citing a hotel reservation system where separating pricing and timezone logic into service classes saved hours of refactoring. The develope…

Alibaba Qwen announced Qwen3.8-Max, a 2.4T-parameter flagship model focused on coding, long-horizon agentic work, and multimodal reasoning, with open weights promised for next week alongside Qwen3.8-27B. The model demons…

Chinese AI startup Moonshot is reportedly seeking greater access to Nvidia's advanced Blackwell GPUs to develop its next-generation AI model, highlighting the growing importance of computing power in the global AI race. …

Genspark has open sourced GenOffice, an AI-native office suite for macOS and Windows, under the Apache License 2.0, covering Docs, Sheets, Slides, and PDF as five Electron apps over one shared engine layer. The suite fea…