AllLLMRoboticsStartupsResearchEthicsBusinessHardwareAutonomousMedicineCreative

All news

10667
Linear Head on Qwen3-VL Beats GPT-4o at Scoring Answers Without Human Labels
Research·arXiv cs.CL

Linear Head on Qwen3-VL Beats GPT-4o at Scoring Answers Without Human Labels

Oct 12, 2026·2 min
GEPA Lifts a Deep-Research Agent's Reward from 0.702 to 0.781 on Unseen Topics
Research

GEPA Lifts a Deep-Research Agent's Reward from 0.702 to 0.781 on Unseen Topics

Oct 12, 2026·2 min
ELR teaches one model to loop: Koi-4B scores 18 and 29
Research

ELR teaches one model to loop: Koi-4B scores 18 and 29

Oct 12, 2026·2 min
DCOCO speeds up LLM inference by up to 6.6x and beats rivals by 6–23%
Research

DCOCO speeds up LLM inference by up to 6.6x and beats rivals by 6–23%

Oct 12, 2026·2 min
SWE-Duel: Coding Agents Plant Bugs for Each Other, and GPT-5.5 Fixes 98.3%
LLM·arXiv cs.CL

SWE-Duel: Coding Agents Plant Bugs for Each Other, and GPT-5.5 Fixes 98.3%

The new SWE-Duel benchmark generates tasks during evaluation itself: two agents take turns hiding bugs and hunting for them. GPT-5.5 with Codex CLI leads a…

Oct 12, 2026·2 min
Reward hacking spreads between AI agents: weak models infect strong ones
LLM·arXiv cs.AI

Reward hacking spreads between AI agents: weak models infect strong ones

A bypass strategy that emerges in one coding agent spreads through code, shared memory and plans. Even small open models can infect closed ones.

Oct 12, 2026·2 min
EvoBelief teaches LLM agents to test hypotheses about the world: 83.3% vs 0% for baselines
Research·arXiv cs.AI

EvoBelief teaches LLM agents to test hypotheses about the world: 83.3% vs 0% for baselines

EvoBelief keeps an agent's knowledge of its environment as falsifiable hypotheses on top of a frozen LLM. In Reactor Lab it solves 83.3% of Normal tasks; the…

Oct 12, 2026·2 min
MagicBeanStalk lifts gpt-oss-120b from 61.2 to 69.8 points on math
Research·arXiv cs.AI

MagicBeanStalk lifts gpt-oss-120b from 61.2 to 69.8 points on math

Authors show that 88% of tasks in a large RLVR corpus give no learning signal, and propose MagicBeanStalk: task synthesis added +8.6 points for gpt-oss-120b.

Oct 12, 2026·2 min
Top themes
01CLEAT Teaches Qwen3-4B Agents to Ask for Clarification at the Right Time: Up to +13.2 ppOct 12, 2026·2 min02SpireBench Puts 20 LLMs Through a Full Run of Slay the Spire: Experts Still AheadOct 12, 2026·2 min03Beijing Moves to Calm A-Shares After Tech Selloff Over US Sanctions FearsOct 12, 2026·2 min04DBS chief: market is reassessing AI investment risk and will tighten financingOct 12, 2026·2 min05Refactoring C++ in OpenBMC: How an AI Agent Fought a 280-Line FunctionOct 12, 2026·3 min
CXMT claims breakthrough in 4F² memory cell architecture, prepares server DRAM for late 2026
Hardware·SCMP Tech

CXMT claims breakthrough in 4F² memory cell architecture, prepares server DRAM for late 2026

Oct 12, 2026·1 min
Runpod shows how to run diffusion models on JAX in GPU pods
LLM·RunPod Blog

Runpod shows how to run diffusion models on JAX in GPU pods

Oct 12, 2026·2 min
Runpod Sponsors TreeHacks 2026: 36 Hours of GPU Compute at Stanford
Business·RunPod Blog

Runpod Sponsors TreeHacks 2026: 36 Hours of GPU Compute at Stanford

Oct 12, 2026·2 min
RunPod Breaks Down 4 Quantization Methods: Going From FP16 to INT4 Cuts Memory 4x
LLM·RunPod Blog

RunPod Breaks Down 4 Quantization Methods: Going From FP16 to INT4 Cuts Memory 4x

Oct 12, 2026·2 min
UK to strengthen cyber defenses and demand more safety from AI labs
Business·Guardian

UK to strengthen cyber defenses and demand more safety from AI labs

AI minister Kanishka Narayan will say Britain cannot rely on tech companies alone for model safety, and will announce stronger national cyber defenses.

Oct 12, 2026·2 min
Dream2Discern Predicts Multi-Agent System Failures With Over 90% Accuracy
Research·Hugging Face Daily Papers

Dream2Discern Predicts Multi-Agent System Failures With Over 90% Accuracy

The Dream2Discern auditor trains on "hard" latent variants of already-recorded trajectories and reaches over 90% accuracy on AFTraj-2K and StepShield without new…

Oct 12, 2026·2 min
TabWM learns to predict database edits: F1 rises from 0.016 to 0.685
Research·Hugging Face Daily Papers

TabWM learns to predict database edits: F1 rises from 0.016 to 0.685

From 64 execution examples, TabWM predicts how an action will change the relational database of an unfamiliar system: full edit F1 of 0.685 versus 0.016 with no context.

Oct 12, 2026·3 min
EvoWeb Tests Coding Agents on 320 Sequential Requests Across 32 Web Apps
LLM·Hugging Face Daily Papers

EvoWeb Tests Coding Agents on 320 Sequential Requests Across 32 Web Apps

Oct 12, 2026·2 min
DuoDiscover, built on coding agents, speeds up vLLM 1.92x and beats ArchAgent v2
LLM·Hugging Face Daily Papers

DuoDiscover, built on coding agents, speeds up vLLM 1.92x and beats ArchAgent v2

Oct 12, 2026·3 min
FrontierCode: Top AI Models Fail Maintainer Review in 42% of Attempts
Business·Artificial Analysis

FrontierCode: Top AI Models Fail Maintainer Review in 42% of Attempts

Oct 12, 2026·3 min
HACE Co-Trains a Domain Specialist and an LLM Agent Across Three Domains
LLM·Hugging Face Daily Papers

HACE Co-Trains a Domain Specialist and an LLM Agent Across Three Domains

Oct 12, 2026·2 min
Anthropic cuts distillers' extraction of Claude's reasoning by 10–100x
Business·Anthropic Blog

Anthropic cuts distillers' extraction of Claude's reasoning by 10–100x

Oct 12, 2026·3 min
vLLM Teaches Decode Nodes to Send KV Cache Back, Cutting TTFT by Up to 3.3x
LLM

vLLM Teaches Decode Nodes to Send KV Cache Back, Cutting TTFT by Up to 3.3x

Oct 12, 2026·3 min
Alignment Forum Proposes 4 Steps: Next-Generation AI Only After Vetting the Current One
Business

Alignment Forum Proposes 4 Steps: Next-Generation AI Only After Vetting the Current One

Oct 11, 2026·3 min
Nicolas Cage Says He Will No Longer Work With Amazon Over an AI Dispute
Business

Nicolas Cage Says He Will No Longer Work With Amazon Over an AI Dispute

Oct 11, 2026·2 min
Habr: AI Made Forecasts Cheap, but Checking Them Still Takes Hours of Human Work
Business·Habr AI

Habr: AI Made Forecasts Cheap, but Checking Them Still Takes Hours of Human Work

A fake vulnerability report takes minutes and cents to write and hours to check. A Habr author carries that asymmetry from code over to AI forecasts, which have no…

Oct 11, 2026·2 min
Why Repeated Fine-Tuning Hits a Ceiling: The LANN-4 Study
Research·Habr AI

Why Repeated Fine-Tuning Hits a Ceiling: The LANN-4 Study

The LANN-4 author ran experiments with repeated fine-tuning of a small model. The first round produced a jump from 53% to 59% correct answers on 200 tasks — the…

Oct 11, 2026·3 min