A new independent evaluation of OpenAI's Privacy Filter (OPF), a 1.5B-parameter PII detector, across 42 benchmarks in 22 languages and 5 domains finds that OPF outperforms Presidio and XLM-RoBERTa on …
Researchers introduced VIVID (Vietnamese Idioms for Validation and Interpretation Depth), the first benchmark for evaluating figurative language understanding in Vietnamese, comprising 1,636 idioms an…
A new arXiv preprint (2608.02665v1) finds that evaluating large language models on a single canonical prompt surface underestimates unsafe compliance by 3.3 to 12.9 percentage points across five model…
Microsoft Research on August 4th spotlighted PRISM2, a multimodal pathology model developed by Paige and Microsoft researchers, trained on 2,350,518 whole-slide images from 685,507 specimens and about…
A developer's colleague incurred a $1,200 bill in a two-hour AI coding session, highlighting the hidden costs of AI assistants. Token consumption, especially for code, is 1.5-2.5x higher than prose, a…
A two-year analysis by Soamee, an AI integration firm, reveals that the real cost of running large language models in production is driven by inflated input token counts from system prompts, conversat…
Kaspersky's report on 33,352 AI-impersonation attacks in Q1–Q2 2026 identifies GigaChat as one of the most commonly spoofed brands, prompting Sber to clarify the only legitimate entry points: the web …
OpenAI built a realtime voice interaction system for GPT-4o in six months, enabling continuous voice conversations with low latency. The system, detailed in a new technical post, uses a pipeline of sp…
A developer compared OpenAI's Realtime API with LiveKit for enterprise voice agents, highlighting trade-offs in latency, cost, and compliance. The analysis shows LiveKit's modular WebRTC pipeline can …
GitHub has introduced a public preview of enterprise teams model policy targeting for Copilot, allowing admins to assign specific AI models to specific teams rather than all-or-nothing org-wide access…
A new arXiv preprint (2607.28636v1) introduces Chain-of-Models (CoM), an automated audit pipeline in which a second model inspects a first model's reasoning trace to reduce cognitive biases in LLM jud…
Microsoft released Agent Governance Toolkit (AGT), an MIT-licensed, Public Preview toolkit that enforces policy, identity, sandboxing, and SRE for autonomous AI agents, covering all 10 OWASP Agentic T…
ChatPlayground AI Unlimited, a platform that consolidates 20+ leading AI models including GPT-4o, Claude Sonnet, Gemini, DeepSeek, Llama, and Perplexity into one dashboard, is available for a one-time…
A developer argues that retrieval-augmented generation (RAG) does not eliminate hallucination in AI systems, citing Stanford's RegLab study showing legal AI tools hallucinate 17-33% of the time. The p…
TextGrad, a framework for automatic 'differentiation' via text feedback from large language models, was published in Nature on 19th March 2025. The framework, which implements backpropagation through …
An AI researcher benchmarked GPT-4o, Claude 3.5 Sonnet, and Llama 3 70B on detecting reentrancy and integer overflow vulnerabilities in a smart contract. Claude 3.5 Sonnet identified both vulnerabilit…
A team built a data-analysis agent that generated business reports but found that self-review failed to catch obvious errors, such as negative monthly sales, because the generator and checker were the…
Few-shot prompting with well-chosen examples outperformed larger models with zero examples in AI translation tests, cutting terminology errors from 9% to 2% on a 500-string test set. The technique, te…
LangChain 0.2.5 and Ollama 0.1.29 enable running local large language models like llama3.1:8b on a laptop without cloud costs, as demonstrated in a tutorial for developers. The setup uses the langchai…
LinkedIn has introduced a "Seems like AI slop" button that lets users flag posts suspected of being generated by artificial intelligence, acknowledging the prevalence of AI-generated content on the pl…