All entries

SFT-GRPO Data Overlap as a Post-Training Hyperparameter for Autoformalization

arXiv:2604.13515v1 Announce Type: new Abstract: Supervised fine-tuning (SFT) followed by Group Relative Policy Optimization (GRPO) is a common post-training recipe. We conduct a controlled ablation ov

safetyarxiv-cs-lg

All Blog X Post Paper YouTube Reddit GitHub

Clear filters

SHARe-KAN: Post-Training Vector Quantization for Cache-Resident KAN Inference

DGX agent

arXiv:2512.15742v2 Announce Type: replace Abstract: Pre-trained Vision Kolmogorov-Arnold Networks (KANs) store a dense B-spline grid on every edge, inflating prediction-head parameter counts by more t

model-releasesarxiv-cs-lg

16 Apr 2026

Shocking result on my pelican benchmark this morning, I got a better pelican from a 21GB local Qwen3.6-35B-A3B running on my laptop than I d…

model-releasesclem-delangue--x

Shocking result on my pelican benchmark this morning, I got a better pelican from a 21GB local Qwen3.6-35B-A3B running on my laptop than I did from the new Opus 4.7! Qwen on the left, Opus on the righ

Sign up today @ http://portal.nousresearch.com/manage-subscription Run `hermes update` and then `hermes model` to configure

Nous Research is promoting their subscription portal and providing instructions for users to update and configure their Hermes model through command-line tools (`hermes update` and `hermes model`). Th

researchnous-research--x

SiLVR: A Simple Language-based Video Reasoning Framework

arXiv:2505.24869v3 Announce Type: replace Abstract: Recent advances in test-time optimization have led to remarkable reasoning capabilities in Large Language Models (LLMs), enabling them to solve high

researcharxiv-cs-cv

Local Ai

Simplicity Prevails: The Emergence of Generalizable AIGI Detection in Visual Foundation Models

arXiv:2602.01738v2 Announce Type: replace Abstract: While specialized detectors for AI-Generated Images (AIGI) achieve near-perfect accuracy on curated benchmarks, they suffer from a dramatic performa

local-aiarxiv-cs-cv

Simulation-Based Optimisation of Batting Order and Bowling Plans in T20 Cricket

arXiv:2604.13861v1 Announce Type: new Abstract: This paper develops a unified Markov Decision Process (MDP) framework for optimising two recurring in-match decisions in T20 cricket namely batting orde

Singularity Avoidance in Inverse Kinematics: A Unified Treatment of Classical and Learning-based Methods

arXiv:2604.13405v1 Announce Type: new Abstract: Singular configurations cause loss of task-space mobility, unbounded joint velocities, and solver divergence in inverse kinematics (IK) for serial manip

researcharxiv-cs-ro

Slash raises 100M at a 1.4B valuation to expand AI-powered banking platform for online businesses

Business banking startup Slash Financial Inc. today announced that it has raised 100 million in new funding on a 1.4 billion valuation to continue product development and build out its artificial inte

industrysiliconangle

SLQ: Bridging Modalities via Shared Latent Queries for Retrieval with Frozen MLLMs

model-releasesarxiv-cs-cv

arXiv:2604.13710v1 Announce Type: new Abstract: Multimodal Large Language Models (MLLMs) exhibit strong reasoning and world knowledge, yet adapting them for retrieval remains challenging. Existing app

Social media polarization during conflict: Insights from an ideological stance dataset on Israel-Palestine Reddit comments

arXiv:2502.00414v2 Announce Type: replace Abstract: In politically sensitive scenarios like wars, social media serves as a platform for polarized discourse and expressions of strong ideological stance

researcharxiv-cs-cl

SocialMirror: Reconstructing 3D Human Interaction Behaviors from Monocular Videos with Semantic and Geometric Guidance

model-releasesarxiv-cs-cv

arXiv:2604.13581v1 Announce Type: new Abstract: Accurately reconstructing human behavior in close-interaction scenarios is crucial for enabling realistic virtual interactions in augmented reality, pre

Soft Q(lambda): A multi-step off-policy method for entropy regularised reinforcement learning using eligibility traces

arXiv:2604.13780v1 Announce Type: new Abstract: Soft Q-learning has emerged as a versatile model-free method for entropy-regularised reinforcement learning, optimising for returns augmented with a pen

safetyarxiv-cs-lg

Solidroad lands $25M to bring AI to customer support interactions

Solidroad Inc., a startup that develops software for use in training and coaching customer service agents, today said it has raised 25 million in an early-stage funding. The Series A round, led by U.K

industrysiliconangle

Some Theoretical Limitations of t-SNE

arXiv:2604.13295v1 Announce Type: new Abstract: t-SNE has gained popularity as a dimension reduction technique, especially for visualizing data. It is well-known that all dimension reduction technique

Sorry for the long wait, everyone! As I said, Qwen is going to keep open-sourcing!

model-releasesclem-delangue--x

Sorry for the long wait, everyone! As I said, Qwen is going to keep open-sourcing! ⚡ Meet Qwen3.6-35B-A3B：Now Open-Source！🚀🚀 A sparse MoE model, 35B total params, 3B active. Apache 2.0 license. 🔥 Agen

Sources: Cerebras plans to make its IPO public as soon as Friday, aiming to raise 3B+ at a 35B+ valuation, a 60% premium to its $22B February valuation (The Information)

The Information: Sources: Cerebras plans to make its IPO public as soon as Friday, aiming to raise 3B+ at a 35B+ valuation, a 60% premium to its $22B February valuation — Chip startup Cerebras is plan

Hardware

Sources: OpenAI agrees to pay Cerebras $20B+ to use its server chips, double the amount previously associated with the deal, and may receive equity in Cerebras (The Information)

The Information: Sources: OpenAI agrees to pay Cerebras $20B+ to use its server chips, double the amount previously associated with the deal, and may receive equity in Cerebras — OpenAI recently struc

hardwaretechmeme

Sources say Ola's AI venture Krutrim's smart assistant Kruti is unavailable to users and the development of its Indian multilingual LLM, Krutrim 3, has stalled (Swathi Moorthy/The Economic Times)

Swathi Moorthy / The Economic Times: Sources say Ola's AI venture Krutrim's smart assistant Kruti is unavailable to users and the development of its Indian multilingual LLM, Krutrim 3, has stalled — W

Sources: SpaceX told employees that it has moved up the scheduled date for when shares awarded to employees become eligible for sale from May to April (Bloomberg)

Bloomberg: Sources: SpaceX told employees that it has moved up the scheduled date for when shares awarded to employees become eligible for sale from May to April — SpaceX has moved up a scheduled vest

Sources: UK paraplegic athlete Claire Lomas was being filmed for Apple Vision Pro series Adventure when she crashed a microlight in Jordan and died in July 2024 (Mark Gurman/Bloomberg)

Mark Gurman / Bloomberg: Sources: UK paraplegic athlete Claire Lomas was being filmed for Apple Vision Pro series Adventure when she crashed a microlight in Jordan and died in July 2024 — Claire Lomas

Sources: Upscale AI, which builds AI networking infrastructure, is in talks to raise 180M to 200M at a $2B valuation, its third funding round in seven months (Bloomberg)

Bloomberg: Sources: Upscale AI, which builds AI networking infrastructure, is in talks to raise 180M to 200M at a 2B valuation, its third funding round in seven months — Artificial intelligence startu

Tutorials

Sparse Goodness: How Selective Measurement Transforms Forward-Forward Learning

arXiv:2604.13081v1 Announce Type: new Abstract: The Forward-Forward (FF) algorithm is a biologically plausible alternative to backpropagation that trains neural networks layer by layer using a local g

tutorialsarxiv-cs-lg

Sparse or Dense? A Mechanistic Estimation of Computation Density in Transformer-based LLMs

arXiv:2601.22795v2 Announce Type: replace Abstract: Transformer-based large language models (LLMs) are comprised of billions of parameters arranged in deep and wide computational graphs. Several studi

researcharxiv-cs-cl

SparseBalance: Load-Balanced Long Context Training with Dynamic Sparse Attention

model-releasesarxiv-cs-lg

arXiv:2604.13847v1 Announce Type: new Abstract: While sparse attention mitigates the computational bottleneck of long-context LLM training, its distributed training process exhibits extreme heterogene

SpatialEvo: Self-Evolving Spatial Intelligence via Deterministic Geometric Environments

model-releasesarxiv-cs-cl

arXiv:2604.14144v1 Announce Type: cross Abstract: Spatial reasoning over three-dimensional scenes is a core capability for embodied intelligence, yet continuous model improvement remains bottlenecked

Spectral Entropy Collapse as an Empirical Signature of Delayed Generalisation in Grokking

model-releasesarxiv-cs-lg

arXiv:2604.13123v1 Announce Type: new Abstract: Grokking -- delayed generalisation long after memorisation -- lacks a predictive mechanistic explanation. We identify the normalised spectral entropy il

Spectral methods: crucial for machine learning, natural for quantum computers?

arXiv:2603.24654v2 Announce Type: replace-cross Abstract: This article presents an argument for why quantum computers could unlock new methods for machine learning. We argue that spectral methods, in

safetyarxiv-cs-lg

Applications

Spectral Thompson sampling

arXiv:2604.13739v1 Announce Type: new Abstract: Thompson Sampling (TS) has attracted a lot of interest due to its good empirical performance, in particular in the computational advertising. Though suc

applicationsarxiv-cs-lg

SPG: Sandwiched Policy Gradient for Masked Diffusion Language Models

arXiv:2510.09541v3 Announce Type: replace Abstract: Diffusion large language models (dLLMs) are emerging as an efficient alternative to autoregressive models due to their ability to decode multiple to

safetyarxiv-cs-cl

SSD-GS: Scattering and Shadow Decomposition for Relightable 3D Gaussian Splatting

arXiv:2604.13333v1 Announce Type: new Abstract: We present SSD-GS, a physically-based relighting framework built upon 3D Gaussian Splatting (3DGS) that achieves high-quality reconstruction and photore

researcharxiv-cs-cv

Stability Principle Underlying Passive Dynamic Walking of Rimless Wheel

arXiv:2604.13530v1 Announce Type: new Abstract: Rimless wheels are known as the simplest model for passive dynamic walking. It is known that the passive gait generated only by gravity effect always be

researcharxiv-cs-ro

Starship Super Heavy Booster, the most powerful moving object ever made by far

The Starship Super Heavy Booster is claimed to be the most powerful moving object ever created, representing a significant advancement in rocket propulsion technology. As the first-stage booster for S

industryelon-musk--x

Applications

Starship V3 booster & ship will be ready for their first test flight in a few weeks

Starship V3 booster & ship will be ready for their first test flight in a few weeks Both Starship and the Super Heavy Booster have successfully completed the static fire tests and are ready to take to

applicationselon-musk--x

Stein Variational Uncertainty-Adaptive Model Predictive Control

model-releasesarxiv-cs-ro

arXiv:2604.01034v2 Announce Type: replace Abstract: We propose a Stein variational distributionally robust controller for nonlinear dynamical systems with latent parametric uncertainty. The method is

Stochastic Trust-Region Methods for Over-parameterized Models

model-releasesarxiv-cs-lg

arXiv:2604.14017v1 Announce Type: cross Abstract: Under interpolation-type assumptions such as the strong growth condition, stochastic optimization methods can attain convergence rates comparable to f

Structure- and Stability-Preserving Learning of Port-Hamiltonian Systems

arXiv:2604.13297v1 Announce Type: cross Abstract: This paper investigates the problem of data-driven modeling of port-Hamiltonian systems while preserving their intrinsic Hamiltonian structure and sta

Applications

SuperGrasp: Single-View Object Grasping via Superquadric Similarity Matching, Evaluation, and Refinement

arXiv:2603.29254v2 Announce Type: replace Abstract: Robotic grasping from single-view observations remains a critical challenge in manipulation. However, existing methods still struggle to generate re

applicationsarxiv-cs-ro

Support Marty, he’s a cool dude

Support Marty, he’s a cool dude 🚨 I am truly honored to have the support of President Donald J. Trump! The stakes this midterm could not be higher. That's why when I started running, I promised I woul

industryelon-musk--x

Supreme Court Justice Clarence Thomas on Wednesday delivered a televised broadside against progressivism, a political philosophy he describe…

Supreme Court Justice Clarence Thomas on Wednesday delivered a televised broadside against progressivism, a political philosophy he described as an existential threat to America and the principles tha

safetyelon-musk--x

Swap Regret Minimization Through Response-Based Approachability

arXiv:2602.06264v2 Announce Type: replace Abstract: We consider the problem of minimizing different notions of swap regret in online optimization. These forms of regret are tightly connected to correl

Syn-TurnTurk: A Synthetic Dataset for Turn-Taking Prediction in Turkish Dialogues

model-releasesarxiv-cs-cl

arXiv:2604.13620v1 Announce Type: new Abstract: Managing natural dialogue timing is a significant challenge for voice-based chatbots. Most current systems usually rely on simple silence detection, whi

Synthesis and Deployment of Maximal Robust Control Barrier Functions through Adversarial Reinforcement Learning

model-releasesarxiv-cs-ro

arXiv:2604.13192v1 Announce Type: cross Abstract: Robust control barrier functions (CBFs) provide a principled mechanism for smooth safety enforcement under worst-case disturbances. However, existing