TSMC's margins are being squeezed by President Donald Trump's push for American-made AI chips, as the Taiwan-based chipmaker has committed $200 billion to U.S. manufacturing since Trump returned to po…
Samsung is in talks to invest up to €1bn in French AI group Mistral at a valuation of €20bn, according to the Financial Times. The South Korean tech giant's potential investment would position Mistral…
A new study from researchers at the University of Zurich finds that reinforcement learning with verifiable rewards surpasses supervised fine-tuning for legal neural machine translation, though enhance…
A new study introduces the Artificial Intelligence for Indian Legal Question Answering (AILQA) system, which uses embedding and generative models including large language models to improve legal quest…
A study from arXiv finds that structured JSON extraction achieves 96% feasibility accuracy in two-agent LLM relays for travel planning, while narrative summarization drops feasibility to 48% despite p…
A new benchmark called Relay-Bench, introduced in arXiv:2607.18438v1, measures large language models' ability to complete multi-domain reasoning chains in a single prompt, with the leading model GPT-5…
Researchers introduce RF-Agent, a framework using textbook-driven knowledge distillation to build language agents for radio-frequency integrated circuit (RFIC) design, creating the first RF-domain rea…
Researchers propose CASE, a framework combining training-time causal alignment and inference-time structural enforcement to improve chain-of-thought faithfulness in large language models. Experiments …
Researchers present SIFT (Self-Improving, Frozen-gate Training), a dynamic classifier service that uses a cheap CPU-bound pipeline with a SPLADE sparse encoder and LightGBM head, escalating low-confid…
A new study from researchers using LLaMA and Mistral models finds that instruction-tuned Transformers encode discourse relations such as causation and antithesis asymmetrically across layers, with ear…
Researchers propose Dual Attention Residuals (DAR), a method that brings multi-stream interaction into historical retrieval for Transformer models through reciprocal cross-stream addressing. Across de…
A new survey from arXiv (2607.19011v1) finds that multimodal humor in memes, cartoons, and comics remains difficult for AI systems because intended meaning depends on non-literal mechanisms, shared cu…
A new study from arXiv finds that large language models (LLMs) produce Context-Sensitive Factual Hallucinations during multi-step reasoning, where the model possesses relevant knowledge but makes fact…
Researchers present AutoJourn, a system for multi-perspective news generation and bias-aware evaluation using large language models, that extracts diverse perspectives from social media, generates bal…
A new training-time framework called Dependency-Aware Intermediate QA Supervision (DAIS) improves final-answer accuracy on complex reasoning tasks by converting filtered teacher rationales into stage-…
Researchers introduced MIRA-Ev, a clinical argument mining benchmark built on Spanish MIR licensing-exam cases, re-annotated by expert clinicians with span-level premises, claims, and directed support…
A new study from arXiv finds that cross-lingual factual inconsistency in large language models can be mitigated at inference time, with persona prompting outperforming more invasive methods like Contr…
Researchers propose RLAES, a unified LLM framework that jointly optimizes automated essay scoring and feedback generation through reinforcement learning, achieving the best scoring performance among L…
Researchers at arXiv identified a failure mode called repetitive copying in long-context large language models, where models extensively copy input text instead of reasoning. They propose GEAR (Ground…
Japanese companies not traditionally associated with technology, including toilet maker Toto, glass fiber manufacturer Nitto Boseki, and MSG producer Ajinomoto, have become major beneficiaries of the …