Baidu executives told investors on its Q2 earnings call that Chinese customers are increasingly seeking domestic AI chips due to constrained supply, boosting confidence in its Kunlunxin chip business.…
Researchers propose DeAR (Decentralized Agentic Reasoning), a framework that replaces centralized agent protocols with autonomous peer-to-peer collaboration, using capability grounding, thought map na…
Researchers propose CASE (Correctness-Axis SElection), a dynamic selection combiner that trains a linear gate on the answer-token hidden state to select the highest-scoring candidate, and introduce de…
Researchers introduced SpeechSense, a dataset for fine-grained speech sentiment analysis that defines an 8-class taxonomy of interpersonal stances detectable through prosodic cues, addressing limitati…
Researchers at an unnamed institution proposed AISA, a framework using large language models for highway construction safety reporting, achieving 75% held-out accuracy in classifying incidents across …
Researchers introduced AVShift, the first German benchmark for authorship verification under distribution shifts, comprising over 150,000 text pairs across three genres and 21 years. Experiments showe…
Researchers introduced Dripper, a lightweight framework for main HTML content extraction that uses small language models for constrained sequence labeling, achieving a throughput of 3.08 pages per sec…
Researchers propose a risk-controlled framework for LLM-as-a-judge evaluation that calibrates uncertainty thresholds on a held-out set to keep the false discovery rate among accepted verdicts below a …
A new survey from arXiv (arXiv:2608.17084v1) organizes research on uncertainty-aware multimodal large language models (MLLMs) around a decision-centered framework, arguing that uncertainty should be e…
A new arXiv study (2608.17810v1) found that subject-matter experts could not ascribe pedagogical meaning to any latent factors derived from large language models (LLMs) on quantitative reasoning asses…
The IOL-AI Challenge, an open-science competition run on unseen problems from the International Linguistics Olympiad (IOL) 2026 Individual Contest, drew 731 submissions from 46 teams under a strict co…
A new arXiv paper (2608.17088v1) proves that the minimum dimensionality required for perfect multilingual embedding spaces grows only logarithmically with the number of languages, showing there is no …
A new arXiv preprint (arXiv:2608.17218v1) introduces a dataset mapping sentences from 150 novel summaries to their source chapters, finding the task unexpectedly difficult for both human and LLM annot…
A new arXiv paper (2608.17153v1) proposes that only AI agents capable of deliberative 'System 2' reasoning should be allowed to access untrusted documents in Retrieval-Augmented Generation (RAG) syste…
Researchers introduced Q-Interference, a classical quantum-inspired attention mechanism for autoregressive language modeling that augments query and key features with amplitude and learned phase, enab…
A new study from arXiv (2608.17168v1) finds that OpenAI GPT 5.4, a top-tier LLM, scores far from ideal in legal reasoning when forecasting European Court of Human Rights cases, producing structurally …
Researchers introduced Polaris, a system that trains a large language model to generate table descriptions optimized for retrieval effectiveness using Direct Preference Optimization (DPO) on existing …
Researchers introduced PTXBench, a benchmark for evaluating and adapting large language models (LLMs) to use architecture-specific PTX for GPU kernel optimization, measuring functional correctness, ta…
A new arXiv preprint (2608.17744v1) reports that fine-tuning three frontier mixture-of-experts models (Alibaba, OpenAI, NVIDIA; 3.6-4.0B active parameters) to reason in Greek produces no measurable ac…
A study from arXiv (2311.06273v1) found that ChatGPT can predict stock market trends for Microsoft and Google based on Twitter sentiment analysis, with a positive link between ChatGPT's evaluations an…