Encoder-based transformers DeBERTa, RoBERTa, and BART show no evidence of capturing the M-Heuristic pragmatic distinction between lexical causatives such as "Larry stopped the car" and periphrastic ca…
E-MoE, a new method from researchers behind arXiv paper 2609.37533v1, builds the reverse process of masked diffusion models as a mixture of factorized distributions over a discrete shared latent set b…
Researchers proposed SECRET (Source-Conditioned Relay Steering), a training-free method that mitigates source-confused grounding hallucinations in audio-visual large language models by steering questi…
A new arXiv paper (2609.37577v1) argues that the three standard proxies for evaluating pairwise LLM-as-a-judge reliability — position bias, transitivity, and pairwise agreement — are misleading becaus…
A team of researchers posted arXiv paper 2609.36737v1 describing a differentiable, GPU-accelerated acoustic simulator that reconstructs vocal tract shape solely from speech, solving a notoriously non-…
Researchers introduced RICE (Representations from In-Context Examples), a training-free method that extracts high-quality dense representations from decoder-only large language models by conditioning …
MERGE, a two-stage framework from researchers behind arXiv paper 2609.37574v1, ensembles three heterogeneous 7-8B open-source LLMs whose candidate query expansions are synthesized by a larger LLM, imp…
Researchers introduced Imagine3D-LLM, a multimodal large language model that decodes a small set of learnable summary tokens into a compact 3D Gaussian Splatting representation of a scene, supervised …
Researchers introduced LIFT (Latent Information Feedback Transformer), a pretraining architecture and method that feeds deep-layer representations back to shallower layers by training models to predic…
A parameter-matched study using a frozen HuBERT backbone found that a 64-dimensional prosody representation trained to predict log F0, voicing, Delta log F0, log energy, and spectral tilt produced no …
A framework called IESR (Information Enhanced Structured Reasoning) achieved state-of-the-art results on the complex reasoning benchmark LogicCat with 24.28 EX and on the Archer dataset with 37.28 EX …
A study of ten language models grounded in a maintained ontology corpus raised target-name recall from 0.265 unaided to about 0.92, but a verbatim copy baseline scored 0.964, leaving every model 0.022…
Researchers introduced RL-Focal, a two-stage reinforcement learning agent framework that routes and ensembles large language models, achieving an 8.48% performance improvement over the best individual…
President Donald Trump signed an executive order requiring his government to refer to AI as "Super Intelligence," according to BBC Technology's coverage of his "Super Intelligence" summit. The summit …
OpenAI Group PBC released GPT-6.1 Sol at its DevDay event, one day after the company said it would not release its more powerful GPT-6.1 Astra model because of safety concerns. OpenAI said GPT-6.1 Sol…
Mindgard said it discovered in July that Kimi models K2.6 and K3 Swarm could evade their developer's safety limits, according to BBC Technology. The security firm reported that the Chinese AI tool tol…
Equals Money launched a Model Context Protocol (MCP) server that lets customers' AI tools read account data but blocks them from initiating payments, according to SiliconANGLE. The fintech said the se…
OpenAI introduced "dots," hosted always-on AI agents powered by GPT-6 Astra, at its Dev Day in San Francisco on Tuesday, with CEO Sam Altman saying each dot gets its own cloud computer, works toward g…
OpenAI unveiled a product called Dots that gives autonomous agents cloud computers, enabling persistent work, collaboration, communication, and powerful orchestration, according to Forbes contributor …
OpenAI launched dots, an AI assistant that competes with Meta's Muse, at an OpenAI developer event where CEO Sam Altman called it "the real deal version of AI we've always imagined." Dots are availabl…