Search: “models”

Delegated Misalignment: How Multi-Agent Structures Amplify LLM Safety Risks

arXiv:2609.27900v1 Announce Type: new Abstract: Large language models (LLMs) are increasingly deployed in multi-agent systems where a principal agent decomposes tasks and delegates them to subordinate

Human

Content type

All Blog X Post Paper YouTube Reddit GitHub

Clear filters

Model Releases

Fine-Tuning LLMs for Translation: General Forgetting Mitigation Does Not Preserve MT-Specific Instruction Following

DGX agent

arXiv:2609.28395v1 Announce Type: new Abstract: Fine-tuning large language models on parallel data improves translation quality but can cause catastrophic forgetting. Mitigation methods are generally

model-releasesarxiv-cs-cl

24 Sept 2026

Gemini 4 is almost ready, says new Google DeepMind chief

model-releasesthe-verge-ai

Google is reportedly nearing the launch of its long awaited Gemini 4 model, after dawdling behind rival developers on flagship AI releases. Speaking with The Information during his first media appeara

Google is a Leader in the 2026 Gartner Magic Quadrant for Container Management

model-releasesgoogle-cloud-ai

We’re excited and proud to share that Gartner has recognized Google as a Leader for the fourth year in a row in the 2026 Gartner® Magic Quadrant™ for Container Management, based on its Completeness of

I built an open-weight alternative to Jev / TypeSafe - introducing OpenJudgement-4B (early preview)

I’m releasing OpenJudgement-4B-Preview, an experimental Qwen-based model fine-tuned on custom datasets for classification, scoring and true/false judgments. It scores answer options directly, and Pyth

I'm trying to post-train AliceAI-Foundation-80B-A3B-Base myself

Just wanted to share with someone - don't have much to report yet. I am interested in this new AliceAI model and have been wanting to make a community impact for a while - and releasing an initial age

Improving LLM-based Autonomous Web Agents with Filtering

arXiv:2609.27770v1 Announce Type: new Abstract: Autonomous web agents, powered by Large Language Models (LLMs), have garnered significant attention for automating various web-based tasks with multi-st

Agents

KITE: KV-Invariant Transformer Expansion for Efficient Agentic LLM Scaling

arXiv:2609.27294v1 Announce Type: cross Abstract: Scaling a language model is not only a question of final quality: the architectural choice determines how much computation is spent during training, p

agentsarxiv-cs-ai

MLPerf Training Introduces Its First LLM Post-Training Benchmark

MLPerf Training v6.1 adds an agentic reinforcement-learning workload that measures how quickly systems can teach a 397-billion-parameter language model to repair real software projects. The post MLPer

model-releasesmlcommons

Power your agents: Gemini 3.8 Live with Live Avatar is now generally available

model-releasesgoogle-cloud-ai

Following our announcement of Gemini 3.8 Live and Gemini 3.8 Live Extended Thinking last week, we are thrilled to share that Gemini 3.8 Live with Live Avatar is now generally available in Gemini Enter

Safety

Reporting Under Pressure: Separating Factual and Tonal Sycophancy in LLM Statistical Analysis

arXiv:2609.27756v1 Announce Type: new Abstract: Large language models are increasingly asked to analyze data and report what the results mean, a task distinct from the belief- or preference-alignment

safetyarxiv-cs-ai

Scribd, Inc. classifies more than 400 million documents with Gemini batch inference on Gemini Enterprise

model-releasesgoogle-cloud-ai

Scribd, Inc. is home to one of the world's largest collections of human-created content. Scribd’s products leverage one of the world's largest collections of human-created content and intelligent tool

Research

Track2Art: Motion-Centric Articulated Object Model Recovery from 2D Point Trackers

arXiv:2609.27675v1 Announce Type: new Abstract: Understanding articulated objects is fundamental for robotic interaction, requiring accurate rigid-part discovery and the recovery of their kinematic re

researcharxiv-cs-cv

AgentBetta: Verification-Driven Adaptive Configuration of an AI Nano-Agent through Selective Expansion and Verified Contraction

arXiv:2609.23512v1 Announce Type: new Abstract: Large language model agents are typically deployed with predefined configurations, although the required model capability, context, tools, permissions,

BigO(Bench): Can LLMs Generate Code with Controlled Time and Space Complexity?

arXiv:2503.15242v3 Announce Type: replace Abstract: We introduce BigO(Bench), a novel coding benchmark designed to evaluate the capabilities of generative language models in understanding and generati

Diffusion Drafts, AR Verifies: Accelerating Document OCR with Self-Speculative Decoding

arXiv:2609.26638v1 Announce Type: new Abstract: Autoregressive OCR vision-language models accurately convert document images into text and structured markup, but require one sequential decoding step p

GGUFs in transformers natively!

Hey there folks! Aritra here from Hugging Face. I wanted to update you all about the latest changes in `transformers`. We now natively support GGUFs (llama cpp quants). You can use it like so: from tr

ImIR: Image-Instruction Tuning for All-in-One Image Restoration

arXiv:2609.25267v1 Announce Type: new Abstract: Degradations vary widely across images, so a practical restoration system has to handle many degradation types with one model. A recent and effective re

Applications

Lightweight Ranking Heads: Accelerating Multi-Task Experimentation in Production Recommender Systems

arXiv:2609.25433v1 Announce Type: new Abstract: Modern production-scale recommender systems rely on complex, multi-task ranking models. Introducing new prediction tasks into these massive systems ofte

applicationsarxiv-cs-lg

Optimizing Denoising Trajectories in dLLMs: A Lightweight Evolutionary Heuristic Approach

arXiv:2609.26052v1 Announce Type: new Abstract: Diffusion Large Language Models (dLLMs) have recently emerged as a promising alternative to conventional Auto-Regressive (AR) Large Language Models (LLM

PhysAI-Bench: A Benchmark for LLM-Based Agentic Decision-Making in Autonomous UAV-Centric Physical AI

arXiv:2609.23695v1 Announce Type: new Abstract: Recent advances in Physical AI have accelerated the use of foundation models in autonomous systems such as unmanned aerial vehicles (UAVs), which must p

QwenImage 2.1 vs QwenImageEdit 2511 (2.1 is not better for every task)

model-releasesr-stablediffusion

First: That the 7GB (int8) 2.1 comes even close to the 20GB (int8) 2511 is amazing, but maybe because of the smaller size some capabilities got lost. For some tasks 2.1 it is much better than 2511, bu

SafetyFlow: An Agent-Flow System for Automated LLM Safety Benchmarking

arXiv:2508.15526v2 Announce Type: replace Abstract: The rapid proliferation of large language models (LLMs) has intensified the requirement for reliable safety evaluation to uncover model vulnerabilit

Total Cost of Agency: Exact Attribution of Memory Injection Cost in Multi-Agent LLM Workflows

arXiv:2609.23790v1 Announce Type: new Abstract: Every node in a multi-agent large language model (LLM) workflow retrieves context from memory and injects it into its prompt, where those injected token

v0.34.4

model-releasesollama-releases

What's Changed server: fix intermittent 'model not found' errors. by @rick-github in #18438 server: apply structured outputs in a single pass on thinking models by @jessegross in #18479 app: avoid Sys

Apollo Restore: A Foundation LLM for Historical Greek Optimized for Fill-in-the-Middle Restoration of Ancient Greek Texts

arXiv:2609.22455v1 Announce Type: new Abstract: We present Apollo Restore, a 24-billion-parameter large language model for restoring lacunae---physical gaps---in fragmentary Ancient Greek texts. Fine-

AVTR-1: Open Stack for Real-Time Interactive Avatars

arXiv:2609.22913v1 Announce Type: new Abstract: Talking-head and dyadic models now achieve real-time inference, yet fast motion generation alone does not produce an interactive conversation. A live sy

CCTU: A Benchmark for Tool Use under Complex Constraints

arXiv:2603.15309v2 Announce Type: replace Abstract: Solving problems through tool use under explicit constraints constitutes a highly challenging yet unavoidable scenario for large language models (LL

Constrained Decoding Eliminates Structural Failures in Small LLMs but Reveals a Scale-Dependent Semantic Gap

arXiv:2609.23742v1 Announce Type: new Abstract: Small open-source large language models (LLMs) in the 0.6B-4B parameter range are increasingly deployed for structured output generation (JSON, function

Cross-Dialect NER for Bangla Regional Dialects Using Leave-One-Dialect-Out Cross-Validation and Explainable AI

arXiv:2609.22536v1 Announce Type: new Abstract: Bangla, the seventh most spoken language in the world, exhibits significant regional dialectal diversity, with dialects such as Barishal, Chattogram, Sy

I asked 4 LLMs to invent words for things only an LLM could understand. Got a dictionary of their own bad habits instead.

You know the thing where you ask a chatbot what's wrong with your plan and it goes 'there are three problems here' and problem three is clearly the runt of the litter? There's a word for that now. An

model-releasesr-chatgpt

MiMo-V2.6-Flash on vLLM: fixes for 'empty responses' with thinking + tools, and a hidden 2,048-token output cap

Some people here say MiMo-V2.6 is bad with tools and are going back to GLM-5.3-Flash. I spent today running MiMo-V2.6-Flash-RL as the backend for an agent harness, on 2× DGX Spark with vLLM, using the

NVIDIA Isaac ROS 5.0 Advances Agentic, Open Source Robotics Development

model-releasesnvidia-blog

To build and deploy sophisticated robotics applications that can perceive, reason and act in dynamic environments, developers need new physical AI models and tools. The ROS open framework is a project

Hardware

Open-Jev Judgments on CallScreenBench: Calibrated One-Pass Scam Screening with a Small Language Model

arXiv:2609.23959v1 Announce Type: new Abstract: Screening a phone call for fraud needs a trustworthy probability after every caller turn, in milliseconds. Jev-style typed decisions promise exactly tha

hardwarearxiv-cs-cl

Safety

Personalized Federated Reinforcement Learning via Model-Agnostic Meta-Learning: Convergence of Exact and Hessian-Free Meta-Policy Gradients

arXiv:2609.22833v1 Announce Type: new Abstract: We study personalized federated reinforcement learning, in which n agents, each acting in its own Markov decision process, collaborate through a server

safetyarxiv-cs-lg

PSD: Pseudo Self-Distillation of Memory Representation Capabilities for LLM Agents

arXiv:2609.23449v1 Announce Type: cross Abstract: Memory systems are becoming a core component of LLM agents, but constructing and maintaining memory remains expensive because it relies on repeated ca

Qwen 3.8 27B at ~3 BPW on an RTX 3060: GSQ vs ByteShape IQ3-XXS 2.88BPW

Someone recommended that I try the ByteShape Qwen 3.8 27B IQ3-XXS GGUF after seeing my previous testing of the GSQ quant. So I did. And the result was… surprisingly bad. For context, I'm running: RTX

Local Ai

Rebekah 1.8b now available

🚀 THE EDGE AI FAMILY HAS ARRIVED. Today we’re opening the doors to a new generation of open-source and open-weight edge AI models from Next GenAi. Meet Rebekah AI 1.8B. 🧠⚡ Small enough to run at the e

local-air-ollama

Research

Toscani-Fourier Distance on Probability Measures: Wasserstein Control, Topological Equivalence on Model Classes, and Duality

arXiv:2609.23163v1 Announce Type: cross Abstract: Comparing probability measures in machine learning trades transport geometry against computational cost: Wasserstein distances encode the geometry of

researcharxiv-cs-lg

VGG16-MCA UNet: Whole-Tumor Segmentation in 2D FLAIR MRI with Decoder-Side Channel Attention

arXiv:2609.23919v1 Announce Type: cross Abstract: Automated brain tumor segmentation supports diagnosis, treatment planning, and monitoring of disease progression, but building models that generalize

VideoGen-Agent: Reinforcing Video Generation Agents

arXiv:2609.24997v1 Announce Type: new Abstract: Recent advances in video generative models have enabled high-fidelity, temporally coherent video generation. However, these models often struggle to sat

WPBench: A Comprehensive Benchmark for Wind Power Forecasting

model-releasesarxiv-cs-lg

arXiv:2609.24444v1 Announce Type: new Abstract: Accurate, reliable, and deployable wind power forecasting is critical for power system dispatch, renewable energy integration, and electricity market op

Co-Evolving Zero-Day Jamming: Adaptive Attack Synthesis and Graph Attention-Based Online Detection

arXiv:2609.21334v1 Announce Type: cross Abstract: Effective evaluation of zero-day jamming detectors requires robust adversarial models. However, existing attack models often assume prior knowledge of

Do Personality-Tuned LLMs Make Better Social Agents?

arXiv:2609.21857v1 Announce Type: cross Abstract: LLMs are increasingly used in social simulations for socially interactive agents and robots, offering more flexibility than rule-based systems. Howeve

Hardware

How Much of a Real Workload Can LLM-Generated GPU Kernels Actually Reach?

arXiv:2609.21058v1 Announce Type: cross Abstract: Language models can now write GPU kernels that outperform PyTorch. We evaluate five model configurations on KernelBench level 1 and find that a fronti

hardwarearxiv-cs-ai

Safety

LIMBO: Learning and Internalizing Model-Free Barrier Objectives for Agile and Safe Whole-Body Control

arXiv:2609.22075v1 Announce Type: new Abstract: Safe whole-body control requires coordinating collision avoidance and balance under high-dimensional, nonlinear dynamics--making safety certificates dif

safetyarxiv-cs-ro

Open source NotebookLM alternative that runs on Ollama

I'm one of the maintainers of SurfSense, so this is self-promotion. I'm posting here because the model-picking part is what I think this sub will have opinions on, and I want real feedback more than u

model-releasesr-ollama

Putting the question before the context took my local Qwen from 89% to 100% on a decision benchmark, and from ~400 ms to ~80 ms

Small, free finding. I use local Qwen models for typed decisions: a state plus a question with fixed allowed answers, and I read the probability of each answer from the logits of one forward pass inst