<?xml version="1.0" encoding="utf-8" standalone="yes"?><rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom"><channel><title>Jibin Rajan Varghese</title><link>https://codejrv.github.io/</link><description>Recent content on Jibin Rajan Varghese</description><generator>Hugo</generator><language>en</language><lastBuildDate>Fri, 02 Oct 2026 00:00:00 +0000</lastBuildDate><atom:link href="https://codejrv.github.io/index.xml" rel="self" type="application/rss+xml"/><item><title>Released MLCommons Jailbreak Benchmark v1.0</title><link>https://codejrv.github.io/news/jailbreak-benchmark-v1/</link><pubDate>Fri, 02 Oct 2026 00:00:00 +0000</pubDate><guid>https://codejrv.github.io/news/jailbreak-benchmark-v1/</guid><description>The first end-to-end MLCommons run of the methodology, with paired baseline and attack evaluation and an average Resilience Gap of 7.57%.</description></item><item><title>Guardrail-agnostic societal bias evaluation for LVLMs</title><link>https://codejrv.github.io/news/lvlm-societal-bias/</link><pubDate>Fri, 11 Sep 2026 00:00:00 +0000</pubDate><guid>https://codejrv.github.io/news/lvlm-societal-bias/</guid><description>We showed how to measure societal bias in LVLMs without asking about the person in the image, so strong guardrails no longer block the evaluation. Presented at ECCV 2026.</description></item><item><title>Context-aware visual redaction with VLMs and inpainting</title><link>https://codejrv.github.io/news/visual-redaction/</link><pubDate>Mon, 15 Jun 2026 00:00:00 +0000</pubDate><guid>https://codejrv.github.io/news/visual-redaction/</guid><description>We built a context-aware visual redaction pipeline: detect sensitive content, let a VLM describe the scene, then inpaint a coherent replacement for images and video.</description></item><item><title>Cosmos 3: an omni world foundation model for Physical AI</title><link>https://codejrv.github.io/news/cosmos-3/</link><pubDate>Tue, 02 Jun 2026 00:00:00 +0000</pubDate><guid>https://codejrv.github.io/news/cosmos-3/</guid><description>A single open world foundation model that reasons across modalities and generates video, audio, and actions for Physical AI.</description></item><item><title>Judging Newsweek’s AI Impact Awards 2026</title><link>https://codejrv.github.io/news/newsweek-ai-impact/</link><pubDate>Thu, 19 Feb 2026 00:00:00 +0000</pubDate><guid>https://codejrv.github.io/news/newsweek-ai-impact/</guid><description>Serving as a judge in Agentic and Physical AI, recognizing teams that put AI to work with measurable impact.</description></item><item><title>A defensible methodology for jailbreak benchmarking</title><link>https://codejrv.github.io/news/jailbreak-methodology/</link><pubDate>Mon, 16 Feb 2026 00:00:00 +0000</pubDate><guid>https://codejrv.github.io/news/jailbreak-methodology/</guid><description>We published MLCommons’ v0.7 methodology: a mechanism-first taxonomy of single-turn jailbreaks so robustness claims can be reproduced, audited, and defended, not just scored.</description></item><item><title>Released AILuminate Jailbreak v0.5: measuring the Resilience Gap</title><link>https://codejrv.github.io/news/ailuminate-jailbreak-v05/</link><pubDate>Wed, 15 Oct 2025 00:00:00 +0000</pubDate><guid>https://codejrv.github.io/news/ailuminate-jailbreak-v05/</guid><description>A draft benchmark that introduces the Resilience Gap: how much an AI system’s safety grade falls when the same hazards are tested under known jailbreak attacks.</description></item><item><title>Released Cosmos Predict 2.5 and Transfer 2.5</title><link>https://codejrv.github.io/news/cosmos-predict-2-5/</link><pubDate>Wed, 08 Oct 2025 00:00:00 +0000</pubDate><guid>https://codejrv.github.io/news/cosmos-predict-2-5/</guid><description>Unified world prediction and high-fidelity multicontrol simulation for AV, robotics, and video analytics.</description></item><item><title>Released AEGIS2.0, a diverse AI safety dataset and risks taxonomy</title><link>https://codejrv.github.io/news/aegis-2/</link><pubDate>Wed, 15 Jan 2025 00:00:00 +0000</pubDate><guid>https://codejrv.github.io/news/aegis-2/</guid><description>A commercially usable AI safety dataset and risks taxonomy for aligning LLM guardrails, published at NAACL 2025.</description></item><item><title>Released NVIDIA Cosmos, a world foundation model platform for Physical AI</title><link>https://codejrv.github.io/news/cosmos-1/</link><pubDate>Tue, 07 Jan 2025 00:00:00 +0000</pubDate><guid>https://codejrv.github.io/news/cosmos-1/</guid><description>An open-source World Foundation Model platform for Physical AI, with diffusion and autoregressive models plus Cosmos Guardrail.</description></item><item><title>10 New Years Resolutions for AI Safety and Security in 2025</title><link>https://codejrv.github.io/writing/post-6/</link><pubDate>Wed, 01 Jan 2025 00:00:00 +0000</pubDate><guid>https://codejrv.github.io/writing/post-6/</guid><description>A curation of 10 areas where work would advance the state-of-art of AI safety, that we all could share a commitment towards working on in 2025.</description></item><item><title>Released AILuminate v1.0, an industry-standard AI risk and reliability benchmark</title><link>https://codejrv.github.io/news/ailuminate-v1/</link><pubDate>Wed, 04 Dec 2024 00:00:00 +0000</pubDate><guid>https://codejrv.github.io/news/ailuminate-v1/</guid><description>A shared, industry-standard way to grade how AI systems handle prompts across 12 hazard categories, from the MLCommons AI Risk and Reliability working group.</description></item><item><title>Re-imagining ancient architecture with AI</title><link>https://codejrv.github.io/writing/post-5/</link><pubDate>Thu, 10 Oct 2024 00:00:00 +0000</pubDate><guid>https://codejrv.github.io/writing/post-5/</guid><description>An experiment in visualizing cities lost to time, from their descriptions in historical or mythologic literature.</description></item><item><title>Can AI assistants reduce India's judicial backlog?</title><link>https://codejrv.github.io/writing/post-2/</link><pubDate>Tue, 25 Jan 2022 00:00:00 +0000</pubDate><guid>https://codejrv.github.io/writing/post-2/</guid><description>The world&amp;rsquo;s largest democracy has a problem. There are 51 million pending lawsuits, some of them pending for over 30 years. Could AI be the solution?</description></item><item><title>Grab and Go Edge-AI for Industry 4.0</title><link>https://codejrv.github.io/writing/post-3/</link><pubDate>Tue, 25 Jan 2022 00:00:00 +0000</pubDate><guid>https://codejrv.github.io/writing/post-3/</guid><description>Companies with legacy automation need to seek out and deploy grab-and-go AI solutions</description></item><item><title>Music-Mapp : The world's first map of music</title><link>https://codejrv.github.io/projects/music-map/</link><pubDate>Tue, 25 Jan 2022 00:00:00 +0000</pubDate><guid>https://codejrv.github.io/projects/music-map/</guid><description>Learn more about the first attempt at producing a visual and interactive map of music at a full song collection scale.</description></item><item><title>Tips and tricks to code 2x faster</title><link>https://codejrv.github.io/writing/post-4/</link><pubDate>Tue, 25 Jan 2022 00:00:00 +0000</pubDate><guid>https://codejrv.github.io/writing/post-4/</guid><description>A collection of time saving tips and tricks I have gathered over the years, that helps me develop code 2X faster than many peers.</description></item><item><title>Disaster management using unmanned aerial vehicles</title><link>https://codejrv.github.io/news/disaster-management-uavs/</link><pubDate>Thu, 21 Jan 2021 00:00:00 +0000</pubDate><guid>https://codejrv.github.io/news/disaster-management-uavs/</guid><description>We published a book chapter on how UAV swarms can cover disaster terrain, dispatch resources, and restore communications so rescue teams can reach survivors in the first hours after a catastrophe.</description></item><item><title>MusicMapp: a deep learning map of music</title><link>https://codejrv.github.io/news/musicmapp/</link><pubDate>Mon, 15 Oct 2018 00:00:00 +0000</pubDate><guid>https://codejrv.github.io/news/musicmapp/</guid><description>We presented MusicMapp at ACM Multimedia 2018: the first large-scale interactive point-cloud of full-length songs, built from a custom Deep CRNN, with personalized exploration and social sharing.</description></item><item><title/><link>https://codejrv.github.io/projects/robust-ml/</link><pubDate>Mon, 01 Jan 0001 00:00:00 +0000</pubDate><guid>https://codejrv.github.io/projects/robust-ml/</guid><description/></item><item><title>Career</title><link>https://codejrv.github.io/career/</link><pubDate>Mon, 01 Jan 0001 00:00:00 +0000</pubDate><guid>https://codejrv.github.io/career/</guid><description/></item><item><title>Research</title><link>https://codejrv.github.io/research/</link><pubDate>Mon, 01 Jan 0001 00:00:00 +0000</pubDate><guid>https://codejrv.github.io/research/</guid><description/></item></channel></rss>