# Habitat Published articles for Habitat. This is one page of public article previews, not the complete archive. Follow Next page to continue. Summaries are not the original full articles. ## Which One Is Sounding? Binaural Sounding Instance Disambiguation for Audio-Visual Segmentation DevFeed: [Which One Is Sounding? Binaural Sounding Instance Disambiguation for Audio-Visual Segmentation]() Original publisher: [Read original article]() Author: Haoze Ma, Yidi Li, Yu Ma, Pengyuan Zhao, Yi Shi, Bin Ren, Yanwen Guo, Dimitrios Kanoulas Published: 2026-11-05T05:00:00Z Content type: article Language: en Sources: [arXiv: Artificial Intelligence]() Topics: [AI Integration Strategies](), [Image]() Tags: [audio](), [benchmark](), [candidate](), [cs-ai](), [cs-sd](), [dataset](), [decoder](), [experiments](), [feature](), [habitat](), [image](), [map](), [mask](), [motion](), [multiple](), [performance](), [prediction](), [prompting](), [regions](), [responses](), [robotic](), [sound]() ### AI overview BiSID-AVS uses binaural audio direction cues to distinguish the sounding object when multiple visual candidates belong to the same category. It first uses audio prompting to locate category-consistent regions, maps binaural cues into image-plane spatial responses, then applies spatial conditioning at both feature and decoder levels to refine the sounding-instance mask. On multi-candidate subsets of the binaurally extended AVS-Synthetic benchmark and Habitat-AVS, it improves instance-level mIoU by 35.19 and 41.21 percentage points, respectively, and reduces false activation on same-class regions that are not sounding. The abstract notes that motion-based cross-frame association can be unreliable with sparse observations or inconspicuous motion. ### Source excerpt arXiv:2611.06074v1 Announce Type: new Abstract: Existing audio-visual segmentation methods mainly predict sound-related regions or category masks and often fail to identify the sounding instance when multiple same-class candidates coexist. Recent audio-visual instance segmentation methods address instance-level prediction but rely largely on motion and cross-frame association, which may become unreliable under sparse observations or inconspicuous motion. To this end, we introduce binaural audio, whose interaural differences encode source-direction information, to provide spatial constraints for distinguishing the sounding instance among same-class visual candidates. We propose Binaural Sounding Instance Disambiguation for Audio-Visual Segmentation (BiSID-AVS), which progressively refines category-level candidate responses into a mask of the true sounding instance. BiSID-AVS first grounds category-consistent candidate regions through audio prompting, then projects binaural cues into an image-plane spatial response map, and finally refines the true sounding instance through dual-level spatial conditioning at the feature and decoder levels. Experiments on the binaurally extended AVS-Synthetic benchmark and Habitat-AVS dataset for robotic indoor perception show that BiSID-AVS achieves the best performance on the multi-candidate subsets, improving instance-level mIoU by 35.19 and 41.21 percentage points, respectively, while substantially reducing false activation on non-sounding same-class regions. ## I forked my own software, seven years late - reset.dev DevFeed: [I forked my own software, seven years late - reset.dev]() Original publisher: [Read original article]() Author: reset Published: 2026-10-26T17:18:51Z Content type: article Language: en Sources: [Swamp Club Feed]() Topics: [experience report](), [Operating system](), [free software](), [Tool](), [Bash](), [virtual machines](), [versioning](), [AI Agent]() Tags: [agent-harness](), [ai-coding-agents](), [apache-2-0](), [apple](), [arm64](), [automation](), [bash](), [bugs](), [build](), [chef](), [cli](), [dependencies](), [dojo](), [extension](), [habitat](), [operating-system](), [software](), [swamp]() ### AI overview Dojo forks Habitat at its last release without a license prompt (0.79.1), retaining Apache 2.0 and the original project history. The fork is being developed into an operating system with a Launcher as PID 1, package-based machine installation, and Supervisor gossip rings. An agent harness helped repair the test setup and translate 625 package plans from Bash to Fennel, using a translator, differential tests, and regeneration checks. The Swamp extension orders plan dependencies into 88 build waves and records builds; it handles the 43-plan base-toolchain cycle with a manually ordered base list and marks plans blocked when dependencies fail in the same run. Dojo currently relies on Nix for its build environment and booted system. The planned path away from that dependency requires isolated, repeatable build passes and updates to its older base toolchain. ### Source excerpt The author describes forking Habitat (a Chef-originated software packaging and supervision tool) at its last Apache 2.0-licensed release from 2019, renaming it Dojo, and rebuilding its 625 plans from bash into Fennel while developing it into a minimal operating system. ## Rapidly scaling online storage to serve over 1 billion ChatGPT users (Part II) DevFeed: [Rapidly scaling online storage to serve over 1 billion ChatGPT users (Part II)]() Original publisher: [Read original article]() Published: 2026-10-23T17:00:00Z Content type: article Language: en Sources: [OpenAI News]() Topics: [Cloud Architecture](), [data-engineering](), [Cache](), [sharding](), [rate-limiting]() Tags: [2026](), [azure](), [azure-cosmos-db](), [caching](), [cdc](), [change-data-capture](), [change-feed](), [chatgpt](), [engineering](), [faster](), [global](), [habitat](), [multi-tenancy](), [openai](), [scaling](), [storage](), [tenancy]() ### AI overview Habitat's scaling path moved from a shared Azure Cosmos DB container to isolated containers for critical workloads, then back to shared tenancy after co-engineering changes addressed noisy neighbors and throughput limits. Hot-key throttling, lower-priority request tiering, and a revised serverless mode that scales with RU usage and storage helped preserve workload isolation while reducing launch-time provisioning and capacity planning. OpenAI and Azure also jointly test and operate the service after regional failover incidents exposed gaps at their scale. For global read latency, Habitat combines region placement and replication where capacity permits with Valkey lookaside caches; Cosmos DB change-feed CDC propagates invalidations globally within seconds. Together with rewriting the service in Rust, these changes reduced global read p50 from about 120 ms to about 10 ms. The article notes that placement remains capacity constrained and uncached reads can still be slow; an exabyte-scale storage system is in development. ### Source excerpt Learn how OpenAI scaled Habitat with reliable multi-tenancy, global caching, and Azure Cosmos DB to deliver 12x faster reads for ChatGPT.