SShortSingh.
Back to feed

Agent Knowledge Base Refuses to Answer Rather Than Guess Unverified Questions

0
·1 views

A DEV Community post describes an agent knowledge base built to return a refusal instead of an answer when no verified source exists. Every item moves through a strict workflow from draft to verified to published, with status transitions guarded against conflicting updates. Only human operators can approve content, and the verifier and publisher must be different people, with the audit trail persisted. Search and ask endpoints serve only published items, and unanswered questions are logged as knowledge gaps for later review. The author frames the refusal as a deliberate feature that keeps agents from leaking unreviewed or withdrawn information.

Read the full story at DEV Community

This is an AI-generated summary. ShortSingh links to the original source for the complete article.

Discussion (0)

Log in to join the discussion and vote.

Log in

Related stories

0
ProgrammingDEV Community ·

Criminals can intercept your calls without a SIM swap, FBI warns

The FBI warns of a method that lets criminals receive a victim's calls without moving the phone number to another device. Attackers can ask a carrier to enable call forwarding or simultaneous ring, so incoming calls, including those reading login codes aloud, reach the criminal's phone. The setup can be done through the carrier or with a short code on a phone the attacker has already accessed. Recommended protections include asking the carrier to block call forwarding and simultaneous ring, and setting a non-default voicemail password. The advisory is FBI IC3 PSA I-041124-PSA, published April 11, 2024.

0
ProgrammingDEV Community ·

Google Cloud's AlloyDB Express Lets Developers Spin Up a Database in Seconds

A developer tested AlloyDB Express, a preview offering from Google Cloud designed as a lightweight version of AlloyDB for PostgreSQL. The service can be provisioned in about two seconds with public DNS and scales compute to zero when idle, with wake-up taking roughly 20 seconds. It targets prototypes, demos, and AI coding agents, and includes AlloyDB AI features such as pgvector and ScaNN vector search even on its smallest tier. Storage is capped at 20 GB, with no high availability, backups, or SLA, and it currently runs only in a handful of regions including us-west1. The author notes a free tier of 50 compute units and 2 GB storage monthly for the first 12 months, with an upgrade path to full AlloyDB for production use.

0
ProgrammingDEV Community ·

Open dataset of 220+ U.S. franchise costs and disclosures released under CC BY 4.0

A new open dataset covering more than 220 major U.S. franchise brands has been released as the 2026 edition under a Creative Commons Attribution 4.0 license. It compiles structured financial figures drawn from audited Franchise Disclosure Documents filed with state registries, including initial investment ranges, franchise fees, royalty and advertising formulas, financial performance representations, and unit counts and turnover rates. The data is available in CSV and JSON formats through a GitHub repository, along with a Python example for filtering brands by cost and royalty rate. The developers say the goal is to replace data previously buried in lengthy regulatory PDFs, aiding software developers, economists, and prospective franchisees. The dataset is positioned as a tool for objectively comparing franchise opportunities without relying on sales materials.

0
ProgrammingDEV Community ·

OGG o MP3: cuándo conviene convertir un archivo de audio y cuándo no

OGG es un contenedor que puede llevar audio Vorbis u Opus, mientras que MP3 es a la vez formato y códec compatible con casi cualquier dispositivo. Convertir suele justificarse cuando el aparato de reproducción, como un coche antiguo o un editor de vídeo, no admite OGG, o para compartir con alguien no técnico. Si el archivo solo se escucha en el móvil o el ordenador, o es un recurso que un juego necesita en OGG, la conversión no aporta nada. Como ambos formatos son con pérdida, la conversión no recupera calidad, por lo que conviene elegir tasas de bits acordes al contenido, con 192 kbps para música y 64–96 kbps en mono para voz. Se recomienda convertir una sola vez desde el original y comprobar que las etiquetas se conservan.