Show HN: Paper Radar – daily ArXiv triage, now runs on Claude Haiku too
Article URL: https://github.com/Eliot5566/JEV-Paper-Radar Comments URL: https://news.ycombinator.com/item?id=50462542 Points: 1 # Comments: 0
Search headlines, summaries and articles to connect the dots.
Article URL: https://github.com/Eliot5566/JEV-Paper-Radar Comments URL: https://news.ycombinator.com/item?id=50462542 Points: 1 # Comments: 0
Article URL: https://github.com/barddoo/tsfm-pg Comments URL: https://news.ycombinator.com/item?id=50460864 Points: 1 # Comments: 1
Article URL: https://github.com/eax/w-isa Comments URL: https://news.ycombinator.com/item?id=50460697 Points: 2 # Comments: 0
This blog post provides scripts and Docker containers for fine-tuning TRELLIS.2, a 3D generation model, using LoRA adapters on footwear data from the Google Scanned Objects (GSO) dataset. The approach focuses on improving geometric accuracy by training only the sparse-structure and shape flow diffusion transformers while freezing the texture stage. The process includes preprocessing with Blender to generate 16 views per object, converting meshes into latents, and training with Hugging Face Jobs or local Docker. Results show significant improvements in F-score, chamfer distance, and normal consistency on held-out test objects, demonstrating effective adaptation for specific 3D object categories.
arXiv:2611.13172v1 Announce Type: new Abstract: Text-to-image diffusion models can render a photorealistic analog clock but cannot control the time it displays. Across four pretrained models, fewer than 1% of generated clocks place both hands within 6 degrees of the requested time; the models instead collapse onto visually dominant configurations such as the canonical 10:10 pose. The same failure appears when a specific fraction of a circular object is requested, where models return complete or conventionally sliced objects. We observe that these two failures share a geometric structure: both target quantities can be parameterized by angular variables. We introduce a geometric token encoding framework that conditions generation directly on these angles. Each angular component is encoded as a periodic sine-cosine feature and projected into a role-specific token embedding that is injected into the prompt before a frozen text encoder. To spatially ground these tokens without requiring localization input at inference, we introduce Stochastic Localization Relaxation (SLR), which stochastically relaxes attention supervision from component-level to object-level to unconstrained localization during training. For clocks, we additionally introduce a hand-swap preference objective that discourages assigning the correct angles to the wrong hand roles. On analog clocks our method reaches 63-78% Acc@6 degrees across evaluation methods, compared with below 1% for the baselines. On circular-fraction generation, it achieves 89.5% Acc@0.10, while maintaining 88.8% on unseen object categories and 84.9% on unseen target values, compared with roughly 20% for pretrained baselines.
On one NVIDIA Vera Rubin NVL72 rack, vLLM serves DeepSeek-V4.1-Flash on SemiAnalysis AgentX at 1.7M tokens/s per GPU at a P90 of 387 tokens/s per user, 58x GB300 NVL72 at the same interactivity.
TL;DR We serve Meta’s Muse Spark Open, an 816B-parameter mixture-of-experts model, on a single TPU v7 host of four chips, with SGLang running on the TorchTPU PyTorch backend. At the... The post Optimizing Muse Spark Open on TPU v7 with SGLang-TorchTPU appeared first on PyTorch.
ComfyUI v0.45.0 introduces multiple updates including new partner nodes for Pika, Tencent, Kling, OpenAI, Qwen, Gemini, Grok, Breeze, and ByteDance; performance enhancements for Qwen 3.x; bug fixes for crashes, database issues, and VAE selection errors; improved memory management with adaptive slice lengths; support for new models like Seedance 2.5, Seedream 5.5, and MiniMax-H3-image; added video input for Qwen 3.x, solid mode in mesh extraction, and texture rebaking; updated workflow templates to v0.11.105 and API documentation; default activation of local_web_sso and partner_node_token; and contributions from new developer GACLove.
On Qualcomm Adreno GPUs, scalar flash attention was limited to 8 rows per workgroup when head size exceeds 128 to prevent register spilling and performance degradation. This change improved performance significantly, reducing execution time from several seconds to milliseconds and enabling completion of the test-backend-ops case.
General, often expert, and sometimes jagged
Article URL: https://www.sandromaglione.com/articles/indie-game-development-with-ai Comments URL: https://news.ycombinator.com/item?id=50432523 Points: 22 # Comments: 16
Article URL: https://www.wired.com/story/meet-the-sad-wives-of-ai/ Comments URL: https://news.ycombinator.com/item?id=50438429 Points: 27 # Comments: 13
NVIDIA Nemotron 3.5 Super VL, a 120B-parameter multimodal vision-language model, is now available in Amazon SageMaker JumpStart. Deploy the NVFP4 or BF16 variant in a few steps, then run text, image, and video inference for document intelligence, computer-use agents, and multimodal agentic workloads.
arXiv:2611.10594v1 Announce Type: new Abstract: Multi-scene video generation must follow scene-specific prompts while preserving consistency across scenes. Dense joint generation enables cross-scene information exchange but incurs quadratic attention cost, whereas separate generation avoids this cost at the expense of cross-scene consistency. We present OneReel, a training-free system that exploits scene structure to enable efficient multi-scene generation. Our key insight is that multi-scene generation can preserve complete intra-scene attention while restricting cross-scene information exchange to a bounded shared memory. OneReel uses scene prompts and durations to establish scene-specific attention scopes and dynamically selects a bounded set of representative latent frames for cross-scene sharing. This reduces video attention complexity from quadratic to linear in the number of scenes. To translate these savings into execution speedups, OneReel packs scene-local and shared K/V into dense variable-length attention and employs work-balanced multi-GPU partitioning with communication-aware selection between K/V gathering and Ulysses-style parallelism. Our evaluations across four video generation models show that OneReel achieves 1.7-4.9x end-to-end speedup over dense attention and 1.2-2.2x over existing sparse-attention advances, while improving text-following accuracy by 34-37 percentage points and maintaining cross-scene consistency.
A researcher’s trawl through publicly available code found private information was at risk across multiple departments.
Hello thereI built a little Chrome extension that analyzes the content elements on a page, scores their "usefulness" and labels them based on that score. What counts as useful can be customized to your preferences and interests, and it learns from your feedback on the scores.Each element gets a badge with its score; hover it for more details.I'm using Jev, so the basic analysis is very fast and very cheap. You need to add your Jev key and run the small local bridge (it's in the repo).For deeper analysis (e.g. a pareto summary) you bring your preferred AI key.Chrome Web Store: https://chromewebstore.google.com/detail/emjbfdcigbfojhdipap... Comments URL: https://news.ycombinator.com/item?id=50435166 Points: 1 # Comments: 1
Hi HN, I decided to move away from Google Contacts and self-host my contacts two months ago. After installing a Radicale server, I was really missing a way to edit my contacts on the desktop. Nextcloud is overkilled for my needs, and InfCloud is the only other web contacts editor but it hasn't had a release in more than 10 years.So I built what was the missing piece in my setup : Natel, a lightweight self-hosted contacts editor for a CardDAV server. The goal is to edit your contacts on the web, be robust enough to not make you lose anything when there's an edit conflict with the server, and be super light in resources on the server. I came up with building a docker image of just a proxy caddy server and a go process that stores your credentials only in-memory, for a compressed image of less than 30MB and it uses less than 20MB of RAM when idle.Tested against four CardDAV servers : Nextcloud, Radicale, Baïkal and Xandikos. I already use it myself and it syncs with my server and DAVx5 back-and-forth, but I consider it still in beta and want to add more testing to prove compatibility for features I don't (yet ?) use. Note that any field that isn't displayed is never lost, it's just written back as-is.Demo: https://demo.natel.app/contacts Website: https://natel.appI wouldn't have done this side-project if I couldn't delegate writing the code to agents. We iterate on architectural decisions together, and once I'm satisfied with the direction, they write the result in a file (find it in docs/DECISIONS.md in the repo), then write the code to meet the expectations and independently review each other's iterations. I drive the project's direction and decisions, but I don't review the code myself. I keep an eye on the tests to know what's covered, and will read and review code when an agent draws my attention to it. The e2e tests were able to find a bug in SoGO, a CardDAV server : https://bugs.sogo.nu/view.php?id=6269 (and I didn't add that server in the compatilibity list yet because that bug breaks one the promises of the app, which is to be warned if we're trying to delete a contact that is modified elsewhere ; the server doesn't correctly enfore the If-Match instruction)As I don't have that many users yet, I welcome feedback on any part of the project but especially the parts that I have not yet tested like Contacts on a Mac ! Comments URL: https://news.ycombinator.com/item?id=50439845 Points: 1 # Comments: 0
The risk of nuclear catastrophe is real, as is the fallout of the China–US domination game (hi there, spies). But no need to panic about Skynet.
arXiv:2611.09951v1 Announce Type: new Abstract: Humans answer complex visual questions by inspecting short moments, updating what they know, and redirecting attention toward missing evidence. Existing video benchmarks, however, often provide the relevant video, evaluate only final answers or isolated localization, or permit textual shortcuts, leaving multi-round visual evidence seeking underexplored. We introduce Video-DCI, a diagnostic benchmark and functionally matched human-agent protocol for Direct Video-Corpus Interaction. It comprises DCI-Loc for temporal evidence localization and DCI-QA for end-to-end discovery, retention, and composition of cross-video evidence. In DCI-QA, videos within each pool are topically related, yet every answer requires hidden temporal evidence from multiple videos. Given only the question and video pool, humans and agents iteratively inspect bounded local windows, retain supporting evidence, and answer with cited temporal spans. Human trajectories provide reference curves for evidence acquisition and use across rounds. Experiments reveal a consistent retrieval-to-grounding gap. On DCI-Loc, frontier VLMs often retrieve and expose the reference moment, yet convert far less of this exposure into precise final citations than the human reference. On DCI-QA, the loss begins earlier and compounds across videos. Ablation studies further confirm that iterative search outperforms one-shot retrieval. Multi-round search is therefore necessary but insufficient: current agents accumulate candidates and observations more readily than precise, jointly sufficient evidence.
arXiv:2611.08500v1 Announce Type: new Abstract: Graph Transformers improve long-range node-to-node communication through global or structure-aware interactions. However, improving cross-node communication does not determine how computations produced at different depths should be combined. Standard residual streams accumulate successive computations with fixed unit coefficients shared across nodes. This fixed aggregation is a residual-interface constraint: once intermediate contributions have been produced, the interface cannot explicitly reweight them by node. We hypothesize that this constraint can create a depth-wise aggregation mismatch when the relative utility of these intermediate contributions varies across nodes and structural contexts. To study this hypothesis, we formalize Graph Transformer residual computation as a causal node-wise history of graph-propagation and feed-forward network (FFN) contributions, making depth aggregation an explicit interface choice. We realize node-adaptive aggregation over this history with Attention Residuals (AttnRes) while preserving the backbone's graph-propagation and auxiliary-state machinery. Block AttnRes reduces the routing cost of long histories relative to Full AttnRes by pooling consecutive contributions. We instantiate the interface in GPS and GRIT and evaluate it on all five ECHO tasks. Both variants achieve lower mean absolute error (MAE) than their independently tuned backbones on every task, and at least one obtains the lowest mean among all compared methods in each task. Routing-only donor-swap interventions support the mismatch hypothesis, while additional ablations examine the key design choices. Together, these results support node-adaptive depth aggregation as a useful complement to cross-node communication in long-range graph reasoning.