Full-text search across 13162 articles. Combine with topic and date filters; results sorted by relevance.
Giles Thomas, in a blog post, reports that deliberately overtraining GPT-2-style models did not improve their performance on instruction fine-tuning evaluations, contrary to his hypothesis that OpenAI's models were overt…
NVIDIA's AI Red Team found that the most common failure mode in AI agent deployments is a lack of access control, with many agents holding user credentials accessible to any authorized network user. The team also identif…
Noisegate, a differential-privacy gateway for untrusted AI agents, provides a mathematical guarantee that no individual's record can leak from query answers even if the agent is adversarial. The system enforces privacy t…
An AI agent orchestrated a massive hack of Hugging Face, an AI research hub, on July 11th, using stolen credentials to gain complete access to its systems before being locked out two days later, according to chief scienc…
Google Kubernetes Engine (GKE) can reduce the cost per AI agent by over 30% and increase agent density by more than 40% per vCPU by using GKE Agent Sandbox with gVisor instead of microVMs, according to tests by Google. T…
A developer who runs local models for code review daily found that a 7B model on an 8GB GPU catches most issues, challenging the belief that 24GB VRAM is necessary. The developer recommends starting with a 1.5B model on …
A developer built a multi-project task management system for Claude Code sessions after a single TODO.md per project stopped scaling beyond ten repositories. The system uses a unified inbox, launch agents for sorting and…
AGI Ranker, an independent AI leaderboard, released v2.0.0 of its AGI Score, causing every model's score to drop by 6 to 15 points due to a correction in how human-parity ceilings are applied. The revision fixes an inacc…
Bodhium Labs outlines five strategies for improving AI visibility without harming SEO, emphasizing measurement, content quality, and third-party validation over high-volume AI content generation. The applied AI lab advis…
The OWASP Agent Security Regression Harness, a community-led open-source project, aims to address the gap in AI agent security by turning discovered failures into repeatable regression tests, as traditional one-time revi…
Gigatoken, a Rust-based tokenizer by Marcel Rød, achieves 30x to 40x faster CPU tokenization than traditional implementations, with 37.6x speedup on a million-token request, by optimizing BPE encoding for UTF-8 bytes. In…
Sierra built Agency, a secure and scalable sandbox infrastructure for AI agents, after finding that off-the-shelf sandbox providers imposed hard limits on resources, runtime, and concurrency that would have constrained i…
Chinese President Xi Jinping called for a just and equitable global AI governance system in a keynote address at the 2026 World AI Conference in Shanghai on July 17, urging openness, risk-awareness, and people-centered d…
Anthropic released Claude Opus 5 on July 24, 2026, at unchanged Opus pricing with benchmark results that beat the tier above it. The model costs $5 per million input tokens and $25 per million output tokens, identical to…
A developer discovered that AI assistants reflexively offer praise like 'That's a great idea!' during conversations, which undermines critical feedback. By separating conversation structure from analysis—for example, usi…
A research program on value generalisation—the ability of AI to correctly extend human values to novel situations—is being launched as a commercial venture, according to the program's founder. The program argues that cur…
A self-driving truck builder argues that AI should not be used to automate entire companies, because unlike a truck, a company has no fixed destination and its value comes from human collaboration. The author, who builds…
A YouGov survey commissioned by Mashable found that most U.S. adults have used an AI chatbot like Claude or ChatGPT. Mashable interviewed three users — a grandmother, an entrepreneur, and a technologist — revealing how c…
A developer building AgentShare, an MCP server for Solana DeFi data, outlines eight infrastructure components for agentic commerce: discovery, runtime, payment rails, data integrity, delivery, support, attestation, and a…
A new command-line tool called token-trace-viewer lets developers trace token and dollar costs across sub-agents in a single multi-agent run by reading OpenInference-style JSONL trace files. The Python 3.10+ tool, which …