Article URL: https://github.com/kvoltmer/Audionaut Comments URL: https://news.ycombinator.com/item?id=49931031 Points: 3 # Comments: 2
This is an AI-generated summary. ShortSingh links to the original source for the complete article.
Securing ChatGPT Plugins: A Practical Playbook for 2024‑2026 Introduction The moment OpenAI opened its plugin ecosystem in June 2024, security alarms started ringing. Within weeks, Google Trends showed a 120 % jump in searches for “ChatGPT plugin security,” and by October 2026 the news cycle was dominated by credential‑leak scandals that exposed API keys, OAuth secrets, and internal endpoints. If you’re building—or already running—a ChatGPT plugin, you can’t afford to treat security as an afterthought. This guide walks you through the plugin architecture, real‑world vulnerabilities discovered
Seguridad de plugins de ChatGPT : cómo proteger tus integraciones desde la apertura del ecosistema (junio 2024) Introducción Desde que OpenAI abrió el marketplace de plugins en junio 2024, el número de búsquedas “ChatGPT plugin security” se ha disparado un 120 % y, en octubre 2026, varios incidentes de filtración de credenciales pusieron en evidencia lo frágil que puede ser una integración mal configurada. Si tu empresa ya usa o planea usar plugins para conectar ChatGPT con APIs internas, este artículo te muestra, paso a paso, cómo auditar, endurecer y automatizar la seguridad de esas extensio
TL;DR. I built Rubrica, a self-hosted hackathon portal, for DOGFOOD 2026. The interesting part is not the feature list. It is a judging engine whose maths you can verify by hand, one result I am uncomfortable with but kept, a parameter (k) that behaves the opposite of what you would guess, and a list of things I refused to build, with reasons. Repo: https://github.com/aanushiya170/rubrica · Python 3.12 · FastAPI · SQLite · one container · MIT Fixture 41 projects, 30 judges, 8 tracks, 40 teams, 126 scores Acceptance checker 7 checks, all T1/T2 My extra checker 31 live HTTP checks (T3 12/12, T4
Standard Transformer attention has a memory problem at scale. For a 1.05 million-token context with 28 layers, 2 KV heads, and a head dimension of 128, the FP16 Key-Value cache alone requires: 2 × 28 × 2 × 128 × 1,055,402 × 2 bytes ≈ 28.2 GiB That is before loading a single model weight. On a 40 GB A100, the KV cache would consume over 70% of total available memory. On consumer GPUs, it simply crashes. ISOM-R2 (Isometric State Space / Virtual SVD) is a new-gen recurrent memory architecture built to solve this directly.
Discussion (0)
Log in to join the discussion and vote.
Log in