A study of Mistral-7B found that demographic identity is readable and faithfully arranged in attention heads, but not causally used, with a single head (L11 H16) faithful across all six attribute type…
A new arXiv paper (2608.18795v1) quantifies why majority voting among LLM samples can backfire on hard questions, introducing a pluralistic agreement index Gamma decomposed into a mechanical preferenc…
A new arXiv paper (arXiv:2608.18816v1) argues that AI hallucinations in large language models have philosophical significance for machine consciousness, presenting two empirical investigations. The au…
Researchers introduced TextQ-German, a dataset suite for human-centered evaluation of German natural language generation from a Quality of Experience perspective, covering text summarization and machi…
Researchers propose SMTrap, a CPU-only framework that uses Satisfiability Modulo Theories (SMT) solver conflict counts to generate inference-heavy Constraint Satisfaction Problem (CSP) instances, enab…
A new arXiv study (2608.18931v1) finds that test-time scaling (TTS) methods for language models fail on open-ended tasks because exploitation—selecting the best output from a candidate pool—is the bot…
Researchers introduced Redakto, an open-source tool for anonymizing text before it is fed to large language models (LLMs), addressing privacy concerns heightened by new EU legislation. The tool, avail…
A new arXiv paper (arXiv:2608.19009v1) proposes Verification Autonomy Levels (VAL), a meta-standard classifying verification schemes for large language models along a single axis: the source of the ve…
A user study of DesCartes Builder, an integrated modeling environment implementing the Function+Data Flow (FDF) domain-specific language, found that the tool achieves good usability for building AI-ba…
Researchers introduced Top-K prompting and the C3LM (Chemistry Constraint-Consistent Language Model), trained on the CREED-CCV-2+USPTO-XL dataset of ~45.6 million verified reactions, achieving state-o…
Researchers introduced BudgetDoc, the first multimodal benchmark for model-budget-performance trade-offs in document tasks, and trained DRB (Document-Reasoning Balancer), a 1B-parameter pre-flight est…
GreekBarRetrieval, a new public benchmark for Greek statutory retrieval derived from GreekBarBench, comprises 283 bar-exam questions and 6,308 candidate statutory articles. Testing three BM25 variants…
Researchers propose MLREF (Module Level Reward Evolution Framework), a new method that uses a module pool to reuse and refine reward components in reinforcement learning, outperforming strong baseline…
Researchers propose rEDMRec, a framework that distills large language model reasoning into four editable experience channels—long-term preference, short-term context, item-perception, and counterfactu…
A new empirical analysis from arXiv (2608.19072v1) finds that large language model (LLM) agents post-training an LLM lock in their training strategy at the very beginning and spend the remaining budge…
Chinese humanoid robot makers are selling machines to government-backed training centres, which then sell the training data back to the manufacturers, according to the Financial Times. This circular a…
The National Republican Senatorial Committee has privately warned top AI companies that the growing public opposition to data centers in Ohio could hurt the GOP's chances in the upcoming election, acc…
OpenAI confirmed that logins and signups on chatgpt.com are failing, with users seeing an authentication key error. The outage affects ChatGPT users attempting to access the service.…
Cognition, maker of the AI coding agent Devin, is in talks to raise funding at a $40 billion valuation after achieving a $1 billion annualized revenue run rate in under two years, according to Bloombe…
Also Inc., a small electric vehicle maker spun out of Rivian, announced it raised $150 million in a Series D round to accelerate its autonomous driving work. The funding will support multiple self-dri…