Search: “reasoning”

In this post, we explore where RFT is most effective, using the GSM8K mathematical reasoning dataset as a concrete example. We then walk through best practices for dataset preparation and reward funct

tutorialsaws-ml-blog

8 Apr 2026

Content type

All Blog X Post Paper YouTube Reddit GitHub

Clear filters

Applications

Seems like a good model from Meta that is still trailing the current series of releases. The most important thing to note is that it is not …

DGX agent

Seems like a good model from Meta that is still trailing the current series of releases. The most important thing to note is that it is not open weights. That was the main reason that Meta's models we

applicationsethan-mollick--x

8 Apr 2026

🇮🇹🇪🇺 This is utterly unacceptable. Reports indicate that Giorgia Meloni is preparing to sideline Roberto Cingolani, CEO of Leonardo, Ita…

🇮🇹🇪🇺 This is utterly unacceptable. Reports indicate that Giorgia Meloni is preparing to sideline Roberto Cingolani, CEO of Leonardo, Italy’s largest defence group. The reason? Multiple sources suggest

researchyann-lecun--x

8 Apr 2026

Writing fiction seems to be a genuine weak spot for LLMs that is not improving as rapidly as almost every other area. There may be a lot of …

model-releasesethan-mollick--x

Writing fiction seems to be a genuine weak spot for LLMs that is not improving as rapidly as almost every other area. There may be a lot of reasons why this is happening. It would be a really interest

8 Apr 2026

Agens Volundr 32B v1: we fixed what you flagged in the Preview. 0/50 loops, session continuation, and an MIT engine that runs it with 262K context (measured to 239K) within a 16 GB budget

Hi r/LocalLLaMA. I'm on the team at Blockway (disclosure: this is our model). Two weeks ago we posted the Volundr 32B Preview here, and you were blunt about what was wrong with it. This is v1, and mos

18 Oct 2026

Fish Audio Drama 3 says it needs a 48 GB NVIDIA card. My 16 GB Radeon disagrees.

Drama 3 officially needs an NVIDIA GPU with 48 GB. I wanted it on my RX 9070, so I built Red Curtain together with Claude. How it works: it's a 35B MoE with ~3B active per token. The FP8 experts (~32

17 Oct 2026

Great paper from NVIDIA sharing an open midtraining recipe for agentic coding. NVIDIA kept SFT identical and added one midtraining stage. SW…

Great paper from NVIDIA sharing an open midtraining recipe for agentic coding. NVIDIA kept SFT identical and added one midtraining stage. SWE-bench Multilingual went from 55.8 to 62.8 on Nemotron 3.5

model-releasesdair-ai--x

17 Oct 2026

A team of Gemini 4 Argon agents beats Minecraft

arXiv:2610.22319v1 Announce Type: new Abstract: We placed four Gemini 4 Argon agents in a sandboxed Minecraft survival world and gave them one goal: defeat the Ender Dragon. Each agent controlled a pl

Applications

Affordance-Aware Interactive Decision-Making and Execution for Ambiguous Instructions

arXiv:2602.05273v2 Announce Type: replace Abstract: Enabling robots to explore and act in unfamiliar environments under ambiguous human instructions by interactively identifying task-relevant objects

applicationsarxiv-cs-ro

AI Consilium: a Benchmark of LLM Clinical Multidisciplinary Second Opinions

arXiv:2610.22298v1 Announce Type: cross Abstract: Large language models (LLMs) can generate differential diagnoses for complex cases, but a useful second opinion must also be factually grounded, clini

Agents

AI Scientists for Building Virtual-Cell Models

arXiv:2508.02276v3 Announce Type: replace-cross Abstract: Autonomous agents, from virtual laboratories to biological data-analysis agents and co-scientists, have begun to turn scientific goals into hy

agentsarxiv-cs-ai

AlignSAE: Concept-Aligned Sparse Autoencoders

arXiv:2512.02004v4 Announce Type: replace-cross Abstract: Large Language Models (LLMs) encode factual knowledge within hidden parametric spaces that are difficult to inspect or control. While Sparse A

researcharxiv-cs-cl

Can Visual Generation Expand What Unified Multimodal Models Know?

arXiv:2610.21459v1 Announce Type: cross Abstract: Unified multimodal models (UMMs) receive image-generation supervision that exposes them to visual information not fully documented in text. Can learni

Compatibility Matters: A Framework for Token-Level Teacher Selection in On-Policy Distillation

arXiv:2610.21234v1 Announce Type: new Abstract: On-policy distillation (OPD) has recently been introduced as a promising paradigm for continual learning, where a student policy can be supervised at th

safetyarxiv-cs-lg

Evaluating Model Capabilities After Benchmark Saturation

arXiv:2610.20667v1 Announce Type: new Abstract: Rapid advances in AI capabilities are causing benchmarks to saturate. Once several models approach perfect scores, a benchmark loses its power to differ

Evolving Together, Failing Apart: The Skill-Policy Co-evolution Trap in Self-Improving Agents

arXiv:2610.20845v1 Announce Type: new Abstract: Skill-policy co-evolution, in which an agent's policy model and its skill bank improve each other during training, is a promising paradigm for self-impr

Four evaluation engineering lessons learned from building agent plugins

model-releasesgoogle-cloud-ai

The real challenge of building an agentic system isn't the construction, it's the measurement. Evaluation engineering is a critical foundation for building reliable AI systems. Setting up systematic,

GraphBEAT: Graph-based Bucketized Event Aggregation Over Time for Early Risk Stratification of Gastrointestinal Cancers

arXiv:2610.22014v1 Announce Type: cross Abstract: Longitudinal electronic health records (EHRs) contain complementary diagnostic and laboratory information about patient health. However, irregularity,

Hierarchical Interpretable Multitask Audio Representation Learning: An Analysis of Event-Scene Relationships through the Information Bottleneck

arXiv:2610.21520v1 Announce Type: cross Abstract: Joint Acoustic Scene Classification (ASC) and Audio Tagging (AT) models traditionally rely on unconstrained multihead architectures. However, this des

ISW-Agent: Integrating LLM-Driven Dialogue with n8n Workflow for Internal Solitary Wave Remote Sensing Image Processing

arXiv:2610.21502v1 Announce Type: new Abstract: Internal solitary wave (ISW) remote sensing requires task parameters and spatial information to be transferred across image acquisition, detection, stri

Learning from Interaction Experience via Localized Context Distillation

arXiv:2610.21427v1 Announce Type: new Abstract: Learning from interaction experience is essential for enabling large language models to continually improve during deployment. Context distillation offe

Learning to Think and Explore in 3D World

model-releasesarxiv-cs-cv

arXiv:2610.21783v1 Announce Type: new Abstract: We study how vision-language models (VLMs) can learn to think and actively explore in 3D space. Instead of providing multi-view images, videos, or 3D re

Applications

LTM: Latent Transition Based In-Context Memory for Vision-Language-Action Models

arXiv:2610.21362v1 Announce Type: cross Abstract: Robotic manipulation often requires remembering interactions that are no longer visible. We introduce Latent Transition Memory (LTM), which stores his

applicationsarxiv-cs-ai

Hardware

Making every stage of medical device document assessment with large language models deterministic

arXiv:2610.21300v1 Announce Type: new Abstract: To show that a medical device complies with regulation, experts judge whether documents and records meet each requirement and record the judgment for ot

hardwarearxiv-cs-cl

Memorilla: Latent Semantic Memory for LLMs

arXiv:2610.21375v1 Announce Type: cross Abstract: Memory enables large language models (LLMs) to use information beyond their parameters, such as past user conversations, long documents, or an agent's

Memory as Weights: Internalizing Long-Term History for Streaming Videos

arXiv:2610.22070v1 Announce Type: cross Abstract: Streaming vision-language models (VLMs) must continuously integrate information from the past while remaining responsive to current observations and u

MetaExp: Learning to Self-Improve through Experience via Verifiable Cross-Task Environment Scaling

arXiv:2610.21492v1 Announce Type: new Abstract: Language-model agents can accumulate interaction feedback, but reliably turning that feedback into reusable experience remains challenging. Existing met

Mind the Delay: Latency-aware Query Routing for Heterogeneous LLM pools

arXiv:2607.18253v2 Announce Type: replace Abstract: Large language model (LLM) query routers typically assign each query to a suitable model that balances response accuracy and monetary cost. Subseque

Applications

Ontology Refinement with Embeddings and LLMs: A Case Study on CASTON

arXiv:2610.21466v1 Announce Type: new Abstract: Ontologies are widely used to represent and organize complex knowledge in domains such as biomedicine and healthcare. However, constructing reliable ont

applicationsarxiv-cs-ai

OS-Helix: Unifying GUI Interaction and Code Execution in Native Computer-Using Agents

arXiv:2610.22234v1 Announce Type: cross Abstract: Real desktop automation is neither purely visual nor purely programmatic. Agents must understand user-facing interfaces while efficiently manipulating

Poison Valleys: Diagnosing Path-Dependent Failure in LLM Code Repair

arXiv:2610.21514v1 Announce Type: cross Abstract: When code-repair agents fail, final tests reveal that a patch is wrong but not which earlier commitment caused it or which software obligation failed.

Probe-and-Resume: When and How to Evaluate Agent Execution

arXiv:2610.21088v1 Announce Type: new Abstract: Test-time search can improve agent performance by exploring alternative actions and using a critic to select promising branches. We study this approach

PROBE: Learning to Audit Policy Compliance in Tool-Using LLM Agents

arXiv:2610.22124v1 Announce Type: new Abstract: Modern agentic systems extend language models with external tools, enabling them to complete complex tasks through multi-turn interactions. As these age

Qwen 27B - sacrificing context in order run better quants, pros and cons.

I was running Unsloth IQ4_XS for a while and while it provided me with a healthy 96K context on 20GB vram, I did notice all the issues it had during draft and all the time and local tokens wasted tryi

Qwen-Audio-Jev: A Unified Interface for Audio Decisions

arXiv:2610.21614v1 Announce Type: cross Abstract: Large audio language models (LALMs) have made substantial progress in audio understanding. However, in high-throughput applications requiring frequent

RDL-Bench: A Benchmark for Task-Aware Question Answering over Relational Databases

model-releasesarxiv-cs-lg

arXiv:2610.21382v1 Announce Type: new Abstract: Relational databases (RDBs) remain the primary infrastructure for storing and managing structured data across enterprise and scientific domains. Natural

RenoLayout: Unified Floorplan Renovation and Furniture Layout for Constraint-Aware Residential Design

model-releasesarxiv-cs-cl

arXiv:2610.21366v1 Announce Type: new Abstract: Residential design aims to transform existing homes into functional living spaces while satisfying diverse design preferences and physical constraints.

Agents

ReviewPilot: A Multi-Agent System for Screening and Data Extraction in Systematic Reviews

arXiv:2610.21042v1 Announce Type: cross Abstract: Systematic literature reviews follow the PRISMA process of screening scholarly works and extracting evidence from included studies. Deep research tool

agentsarxiv-cs-ai

RTX 2080 Ti 22GB modded card for local LLM inference – anyone tested one?

Title: I'm building a small local AI server to validate a project I've developed. The application currently runs Qwen 9B and the server will only be used for an initial pilot with a few users. I have

UniFGVC: Universal Training-Free Few-Shot Fine-Grained Vision Classification via Attribute-Aware Multimodal Retrieval

arXiv:2508.04136v4 Announce Type: replace-cross Abstract: Few-shot fine-grained visual classification (FGVC) aims to discriminate between subtly different categories with very limited data. While Mult

V-SHARE: Provable Convergence Guarantees for Advantage-based Algorithms

arXiv:2610.21448v1 Announce Type: cross Abstract: In reinforcement learning, advantage-based value-learning methods, unlike standard Q-learning, share information across actions by decomposing the act

Waypoint-OPD: Transferring RL-Induced Policy Shifts via Gated On-Policy Distillation

arXiv:2610.21601v1 Announce Type: cross Abstract: Policy-shift distillation reuses the difference between a teacher's log probabilities before and after reinforcement learning (RL) as supervision on s

safetyarxiv-cs-cl

When Reality Outpaces Cognition: A New Scaling Principle for Reliable Intelligence

arXiv:2610.20764v1 Announce Type: new Abstract: This paper challenges the understanding capabilities of existing AI paradigms and emerging world models by asking a fundamental question: How quickly ca

5/5 Across the board, results show the business impact and differentiation of sovereign specialized intelligence, and the opportunity for co…

model-releasesai21-labs--x

5/5 Across the board, results show the business impact and differentiation of sovereign specialized intelligence, and the opportunity for companies to reach AI independence. Thanks to @nvidia, @Gong_i

ACED-Bench: Evaluating How LLM Agents Acquire and Act on Causal Evidence

arXiv:2610.14381v1 Announce Type: new Abstract: Large language models are increasingly used as scientific assistants, yet we still lack reliable ways to evaluate whether they can investigate a system

ActICL: Post-Training for Agentic Multimodal In-Context Learning

model-releasesarxiv-cs-cl

arXiv:2610.19475v1 Announce Type: cross Abstract: Multimodal in-context learning (ICL) is an important capability of multimodal large language models (MLLMs). Its performance is closely tied to the de

ActiveEvidence: Decision-Focused Evidence Graphs with Verifiable Acquisition Gaps

arXiv:2610.20532v1 Announce Type: cross Abstract: A multimodal agent must decide whether visual evidence is worth acquiring, which candidate or resolution to query, and when further interaction can no

safetyarxiv-cs-cv

AlignPoison: Conditional Targeted Poisoning of Generative Recommenders via Decoding-Aligned Prefixes

arXiv:2610.14487v1 Announce Type: new Abstract: Generative recommendation models based on autoregressive decoding have been adopted as alternatives to classical ranking-based systems. While their mode