Knowledge catalogue
arXiv:2609.11133v1 Announce Type: new Abstract: Datacenter GPU power is the binding constraint on LLM serving capacity, and production serving has shifted to prefill/decode (PD) disaggregation. Deploy
arXiv:2609.11347v1 Announce Type: new Abstract: Knowledge graph foundation models such as ULTRA achieve zero-shot link prediction on unseen graphs through dedicated architectures that hard-code a tran
i've been thinking about this more after trying different agent setups. the model isn't the only thing that determines how well an agent performs. The harness around the model matters a lot too. With
arXiv:2609.06853v1 Announce Type: cross Abstract: Shared key--value (KV) cache reuse improves large language model (LLM) serving, but it can also create a timing side channel that reveals whether a pr
arXiv:2609.10286v1 Announce Type: new Abstract: Humanoid locomotion on granular terrain remains a significant challenge due to its complex foot-terrain interaction dynamics that are difficult to model
arXiv:2609.07008v1 Announce Type: new Abstract: Multi-head latent attention (MLA) exposes many logical query heads through one packed latent KV stream. This representation is memory efficient, but it
Bloomberg: Sources: Sam Altman told OpenAI employees that the company is considering slowing cutting-edge AI development, and he hopes other AI companies will do the same — OpenAI is considering slowi
arXiv:2609.08307v1 Announce Type: cross Abstract: Large language models (LLMs) are increasingly used as backends for intelligent web services, but serving them across the edge continuum requires balan
arXiv:2609.08914v1 Announce Type: new Abstract: Pixel-level annotation of fixed traffic-camera imagery is expensive, while crosswalk models trained from street-level imagery face a substantial viewpoi
arXiv:2609.06046v1 Announce Type: cross Abstract: We present a modular, high-fidelity simulation framework for the development and benchmarking of flight control strategies in fixed-wing aerial robots
If they truly believe the risk is that high, and I think they do, then we need 100x more research and understanding of it. That requires them to openly share their models, datasets, training code, and
arXiv:2609.05807v1 Announce Type: new Abstract: Optical Chemical Structure Recognition (OCSR) is a fundamental component of chemical literature mining, enabling molecular database construction, reacti
arXiv:2609.06635v1 Announce Type: cross Abstract: An image-augmentation pipeline must deliver a complete batch before a model can use it. We compare seven input paths from five libraries, starting wit
arXiv:2609.06712v1 Announce Type: new Abstract: Diffusion Transformers (DiTs) achieve strong video generation quality, but their dense spatiotemporal self-attention scales quadratically with sequence
arXiv:2609.07740v1 Announce Type: cross Abstract: Kolmogorov-Arnold Networks (KANs) replace the fixed activation functions and linear weights of Multi-Layer Perceptrons (MLPs) with learnable univariat
arXiv:2609.06144v1 Announce Type: new Abstract: Visual dubbing must synchronize mouth motion with replacement speech while preserving identity, appearance, and temporal consistency. Although X-Dub pro
Executive Summary Since the release of our May 2026 report detailing adversarial misuse of artificial intelligence (AI), Google Threat Intelligence Group (GTIG) has observed forward leaning adversarie
French artificial intelligence lab Mistral AI SAS today announced that it has raised €3 billion, or about $3.49 billion, in funding. Samsung Electronics Co. Ltd. led the investment. It was joined by m
Other investors in the round include Benchmark, Bessemer, Kleiner Perkins, Greylock, Lightspeed, Altimeter, Bond Capital, Meritech, Atreides, Valor, T. Rowe Price, Lux, 8VC, D1, 137 Ventures, DST, Pos
Hey guys, After my CPU-only to 96GB VRAM test, I tested Qwen3.8-Flash-Next across llama.cpp, SGLang and FreeToken on the same workstation. This time I wanted to see what changes when you keep the hard
arXiv:2609.04557v1 Announce Type: cross Abstract: Quantum error-correction decoders are typically benchmarked against synthetic circuit-level noise, under the assumption that a decoder's ranking under
I moght get shit on for posting this but, I feel like i don't see this being talked enough and it feels like such a waste of a good piece of software. Exl3 is incredible, albeit only if you have NVIDI
I ran a 112-case MiniMax H3 quality sweep on an RTX 3090 last night trying to find the best combos . Kept the prompt, seed, reference image, RefMods, Sage setup, sampler, and scheduler stayed constant
LayerStoRm: GLM-5.3-Flash UD-Q4_K_XL (186 GiB) at 1M context on 2× RTX 5090 + 2× RTX 5080 (96 GB VRAM total) using RAM for the pinned experts. LayerStoRm is a (still experimental) MIT-licensed continu
Hi, I use hermes as a harness, and I am pleased with it, but sometimes Hermes's context size is a bit too much for my system, so I wanted to delegate or straight up use a lighter option for coding tas
arXiv:2404.07124v2 Announce Type: replace-cross Abstract: This paper presents a pipeline designed to bring ultrasound (US) plane pose estimation closer to clinical use, demonstrating the feasibility o
TL;DR: switching KV cache to f16 may give a boost in speed if using MTP and ngrams. I have a self-built 'AI mega-cluster' with 2x P40s on a cheap Chinese motherboard and a Xeon CPU (around $1,100 to b
Hello everyone. I felt current LLM benchmark harnesses hand you headline numbers but offer no tooling to see how models actually answered each question (they dump everything to a JSONL or Parquet file
Hi everyone I'm running four rtx 4090, 64GB ram, on a threadripper pro motherboard so all PCIe x16 ports, as a homelab machine for coding. I was migrating from vllm + qwen 3.8 27B (fp8+256k kv cache)
I just want to use paperless-ai and be able to ask questions relative to it. Bonus points if I could use it with home assistant but that's not the focus. I just can't see needing a 32 GB VRAM GPU for
I've been running Qwen3.8-27B as a local inference server for a production content intelligence pipeline (HVAC industry stuff, lots of long-context retrieval and structured extraction). I have been wa
arXiv:2609.03151v1 Announce Type: cross Abstract: Long-context reasoning for large language models (LLMs) is becoming increasingly important, but training over long sequences remains challenging due t
arXiv:2609.03557v1 Announce Type: new Abstract: Action-conditioned video models require large-scale visual data paired with control signals that are temporally aligned with the resulting scene transit
arXiv:2609.04070v1 Announce Type: new Abstract: Bridging the gap between the discrete reasoning of Vision-Language Models and the continuous, physics-constrained nature of autonomous driving remains a
arXiv:2609.03352v1 Announce Type: cross Abstract: Constant optimization refines the numerical coefficients of candidate expressions in tree-based genetic programming for symbolic regression. But its p
arXiv:2609.04105v1 Announce Type: new Abstract: Blackwell's 4-bit floating-point (FP4) tensor cores do not automatically make attention faster because softmax conversion and on-chip dependencies domin
arXiv:2609.03181v1 Announce Type: new Abstract: We present Jina-OCR-v1, an end-to-end document parsing model built to serve on low-budget GPUs. It combines the compressed-vision encoder and the 3B mix
This is a follow-up to my post from yesterday (17 -> 25-29 t/s with the expert cache PR). Same box: 2x RTX 3090 on PCIe 3.0, dual Xeon E5-2696 v4, 188 GB DDR4-2133 LRDIMM, llama.cpp, full 261k context
arXiv:2609.01792v1 Announce Type: cross Abstract: Passive acoustic monitoring of killer whales is particularly important for conservation of the endangered Southern Resident killer whale population, b
arXiv:2511.12018v2 Announce Type: replace Abstract: Traffic safety analysis at signalized intersections is essential for reducing vehicle and pedestrian collisions, yet traditional crash-based studies
arXiv:2609.02319v1 Announce Type: new Abstract: Ultra-low-altitude unmanned aerial vehicles (UAVs) require surround vision near buildings, vegetation, and other obstacles. We present a parallax-aware
For the ninth consecutive year, Gartner® has named Google a Leader in the Gartner Magic Quadrant™ for Strategic Cloud Platform Services, positioned furthest for Completeness of Vision. We believe this
arXiv:2609.02876v1 Announce Type: cross Abstract: Ordinary differential equations (ODEs) underlie models in science and engineering, and many applications need derivatives of their solutions with resp
arXiv:2609.01657v1 Announce Type: cross Abstract: Multimodal models often build on architectures designed for generative vision-language modeling, typically combining separately pretrained vision enco
This is very true, and something we still try to do for all products (and even features) we launch. Find a small group of customers who we think will push us the most on product and then build for the
There is a video stream running on my desktop right now. It has sound, it has never repeated itself, and it will not stop. I point VLC at a local URL and it plays. One RTX 5090 does all of it — no clo
Samantha Kelly / Bloomberg: Robotics and AI startup Lyte, founded by ex-Face ID engineers, raised 165M led by Maverick Silicon at a 1.6B valuation, taking its total funding to $272M — Lyte, a fast-gro
Hi, everyone, I’ve just released VoxGen, a lightweight native inference engine for VoxCPM2, written in Rust and using Vulkan compute instead of Python/PyTorch/CUDA. Why VoxGen? The main reason I start
I'm planning to upgrade my workstation (linux with 5700X/64GB DDR4) for local inference and pytorch training. I'm trying to decide between: 2× AMD Radeon AI PRO R9700 32GB 2× AMD Radeon PRO W7800 48GB
arXiv:2609.00309v1 Announce Type: cross Abstract: AI compute verification is one of the first tangible and tractable points for international policy aimed at AI governance. Determining whether frontie
Not written by AI all mistakes mine. I saw people on the subreddit saying that 3.6 works better without thinking. It made me want to know for certain about which is better, 3.6 or 3.8 for low thinking
arXiv:2609.00188v1 Announce Type: new Abstract: Robotic manipulation faces a fundamental scaling challenge: robust generalization demands broad physical experience, yet action-labeled robot trajectori
arXiv:2608.30105v1 Announce Type: cross Abstract: Deep learning-based side-channel analysis has historically focused on single-byte targets and manually cropped traces, which risks discarding exploita
arXiv:2608.28615v1 Announce Type: cross Abstract: Synthetic personas based on large language models (LLMs) are increasingly proposed as substitutes for human survey respondents, yet systematic validat
More new massive updates from turboderp: - CPU offload of MoE experts - Qwen-3.8-Flash-Next ngram disk offload - GLM-5.3-Flash - New self-calibrated optimization technique - Countless other optimizati
arXiv:2608.30400v1 Announce Type: new Abstract: High-dynamic-range (HDR) images, with their rich tone and detail reproduction, hold significant potential to enhance computer vision systems, particular
arXiv:2608.31074v1 Announce Type: cross Abstract: We propose a lightweight two-stage framework for real-time video anomaly detection. The first stage employs YOLO v11n-pose to detect persons and extra
Super interesting paper from Meta. Long-horizon research agents are coming. But one common problem with research agents today is the lack of originality and how to decide what experiments are worth ex
arXiv:2608.28044v1 Announce Type: cross Abstract: Large language model (LLM) inference serving is priced by tokens, but GPU energy is consumed over inference windows. This accounting mismatch makes to
Hey guys, I tested Qwen3.8 Flash with llama.cpp from CPU-only to the full 96GB of my RTX PRO 6000. Short version: CPU-only reached 8.34 tok/s at a 2K prompt Full 96GB reached 109.07 tok/s At 245K cont