Article URL: https://www.sciencedaily.com/releases/2026/09/260924231343.htm Comments URL: https://news.ycombinator.com/item?id=49856971 Points: 3 # Comments: 0
This is an AI-generated summary. ShortSingh links to the original source for the complete article.
Article URL: https://marcosmagueta.com/blog/a-roman-name-for-software/ Comments URL: https://news.ycombinator.com/item?id=49857173 Points: 4 # Comments: 0
A developer has launched two browser-based applications, CHARFORGE and a 3D Game Engine, running entirely on a budget VPS with 6 vCPUs, 11 GB of RAM, and no GPU. CHARFORGE generates realistic 3D human characters with AI-assigned names, backstories, and clothing, using only locally hosted AI models and no third-party APIs. The companion 3D Game Engine allows users to place forged characters into procedurally generated environments and play in first or third person directly in the browser. To work around the lack of GPU, all 3D rendering and mesh deformation is handled client-side via Three.js, while the server handles only file storage and AI inference. A shared 163-bone skeleton system allows characters created in CHARFORGE to be used seamlessly in the game engine without any retargeting.
Exposed credentials in public code surged 81% to 1.27 million last year, driven largely by AI-assisted development spreading secrets across source code, CI/CD pipelines, and AI tool configuration files. GitGuardian found that 64% of secrets it confirmed valid in 2022 were still unrevoked as of January 2026, highlighting how long leaked credentials remain exploitable. To address the growing triage burden, GitGuardian has added two AI agents to its Public Secrets Monitoring product, covering incidents across GitHub and Docker Hub. A triage agent performs an initial assessment of each incident, escalating stronger candidates to a deeper analysis agent that returns an organizational verdict, a risk score, and visible reasoning. New workspaces receive the feature by default, with a gradual rollout to existing ones, while human reviewers retain final oversight of every incident.
Retrieval-augmented generation (RAG) systems can surface outdated documents that match a query but are no longer approved for current use, posing a reliability risk in AI-generated responses. A freshness gate addresses this by checking retrieved documents against three criteria: whether the revision is currently approved, whether the document is already in effect, and whether its review deadline has passed. The gate is implemented in Python 3.12 using only the standard library, with timezone-aware UTC timestamps to ensure reproducible and consistent evaluation. Authority over which revision is approved must come from an external revision register, not inferred by the model from filenames or document text. This approach narrows the gate to a specific policy use case — answering questions about currently approved material — and does not validate factual accuracy.
Discussion (0)
Log in to join the discussion and vote.
Log in