Analyzes how gradient interactions between racial identity terms and hate speech signals inside SSD architectures shape biased or spurious semantic representations.
Exploits unnatural social gaze patterns in AI-generated images—where eye contact and attention cues are semantically inconsistent—as a detection signal.
> Not a public base model release, but a **model‑enabled security offering** relevant to the “frontier + paradigm shift” criterion.
Applies self-ensembling across multiple vision-language model inference passes to improve structured data extraction accuracy from chart images.
Identifies conditions under which incorporating annotator demographic information improves hate speech detection, characterizing the interaction between data size and modeling choice.
Aligns text representations using contrastive learning to improve semantic matching across variable-length or domain-shifted text pairs.
Translates natural language queries directly into configuration parameters for retrieval-augmented agents, eliminating manual configuration specification.
User correction messages in chat are automatically converted into LoRA fine-tuning training examples to align models with user intent.
A set of 17 specialized capabilities is defined for AI agents operating across social media platforms.
- **ExecuTorch** (deployment framework) – Meta[5] - **React Native ExecuTorch** – Software Mansion, building on Meta’s ExecuTorch[5]
Builds a specialized legal AI assistant that automates competition law research tasks such as case retrieval, analysis, and regulatory reasoning.
Augments a vision-language model with extracted edge maps to improve flowchart understanding for industrial requirements engineering workflows.
GENESIS deploys AI agents to autonomously synthesize, research, and test 6G Radio Access Network configurations without human intervention.
Analyzes how widespread adoption of the same algorithmic hiring tools across employers reduces candidate diversity and concentrates labor market power.
A Windows GUI application simplifies installation, configuration, and management of llama.cpp running inside WSL or Ubuntu.
Combining outsourced services with locally-run AI models is projected to undercut the cost of using frontier lab APIs for many use cases.
Anthropic releases Claude Opus 4.7 and Sonnet 4.6, frontier-class models supporting up to one million tokens of context.
Ranks intrusion detection alerts by risk using subnormal Gaussian fuzzy models to reduce analyst alert fatigue while prioritizing high-severity threats.
Extends two-level Answer Set Programming with quantifiers and weak constraints, establishing complexity results and an efficient solver implementation.
The creator of the LAGK AI governance framework answered community questions, with debate centering on contested framework design choices.
A community-curated collection where users share and discover reusable prompt templates for ChatGPT and other large language models.
An Ontology Anchor mechanism encodes a personalized map of user values and priorities to guide AI behavior and responses.
President Trump appoints Pam Bondi to a White House panel overseeing artificial intelligence policy.
A pull request adds tokenizer support for MiniCPM5 models to the llama.cpp inference library.
A news item references or responds to a quote attributed to entrepreneur and essayist Paul Graham.
Argues that expanded government programs and social safety nets are the appropriate policy response to AI-driven job displacement.