845 articles · Page 13 / 36
How prompt injection rides untrusted content into tool calls, and how to mitigate it with least privilege, sandboxing, fixed schemas, and output validation.
Avoid model-name anchoring by defining success criteria, output format, and failure handling, then running evals on every change.
Overview of EU DSM TDM exceptions and US Copyright Office guidance on AI training, focusing on lawful access and human contribution.
OpenAI’s GABRIEL converts qualitative text and images into measurable outputs, adding reproducible runs, batching, retries, and audit trails.
Seedance 2.0 backlash signals AI video risks shifting from training data to outputs, deepfakes, and distribution controls.
Break coding agent latency into output, prefill, tool time, and network overhead to measure end-to-end duration.
TechCrunch says Codex Spark inference runs on Cerebras WSE-3, highlighting serving bottlenecks and PoC latency metrics.
Design an ops loop to detect provider doc changes and respond using 429 signals, headers, runbooks, and fallbacks.
Practical checklist to reduce citation hallucinations in long-form RAG by auditing chunking, retrieval/reranking, and refusal when evidence is thin.
Explains agentic coding and video generation as iteration-loop gains, emphasizing sandbox control, logs/tests, and evaluation checklists.
How agent link-opening expands the attack surface, and how instruction hierarchy, URL constraints, and sandboxing reduce leakage and injection.
A curated link roundup from recently collected official updates and tech news.
Android 17 reports highlight Secure Lock Device, intrusion logging, and Identity Check expansion—reshaping lock as an OS-level security state.
LLM choice increasingly hinges on structured output, tool calling, caching/batching, rate limits, and data governance—not benchmarks.
Claude Code introduces an agentic CLI loop with shell and filesystem access, shifting development toward permissions, verification, and review.
Cloudflare’s “Markdown for Agents” converts requested HTML pages to Markdown, easing RAG inputs while raising citation, control, and injection risks.
Reasoning vs instant modes trade quality, latency, and cost. Use If/Then defaults, streaming, and progress cues to keep user trust.
How GRPO-style relative ranking and multi-reward signals (format, tool calls, efficiency) shape agentic RL gains and risks in GPT-OSS.
OpenAI dissolved the Mission Alignment team; watch how safety ownership, RACI paths, and SSC/DSB governance appear in upcoming releases.
OpenAI Codex reportedly runs on Cerebras WSE-3, highlighting lower TTFT and reduced round-trip overhead for faster agent UX.
OpenAI shares scaling PostgreSQL to millions of QPS using replicas, caching, rate limiting, and workload isolation to protect DB paths.
Prism, a free LaTeX-native workspace, embeds GPT-5.2 to unify writing, collaboration, and reasoning with a verification-focused workflow.
PersonaPlex combines text role prompts and audio voice prompts to keep consistent personas in low-latency, full-duplex speech conversations.
ZDNET tests six popular AIs with trick questions, highlighting hallucination risk and why teams need RAG, CoT, self-checks, and evaluation rules.