Zhemal Khamidun
Hamidun News editorial desk led by Editor-in-Chief Zhemal Khamidun (CPO of AlpinaGPT). Daily curation across 50+ global AI labs, research feeds, and industry reports in 6 languages.
Zebra3D speeds up 3D reconstruction by up to 7.1× on 1024 frames without training from scratch
Zebra3D's hybrid attention speeds up off-the-shelf VGGT, HunyuanWorld-Mirror and π³ by up to 7.1× on 1024 views. At 94% sparsity, VGGT accuracy drops by only 1.1 pp.
SynAct Teaches LLM Agents to Ask for Help Where It Changes the Outcome: +11.7 pp
SynAct spends expert attention on individual agent decisions, not whole tasks, and raises accuracy by an average of 11.7 pp in 16 of 16 test configurations.
STAPLE adds billions of parameters to OLMo-2 and decodes twice as fast as MoE
STAPLE embeds static sparse memory into a pretrained LLM. On OLMo-2 1B it gains +4.27 points on HumanEval, while training and generation slow by no more than 3%.
SWE-Duel: Coding Agents Plant Bugs for Each Other, and GPT-5.5 Fixes 98.3%
Reward hacking spreads between AI agents: weak models infect strong ones
EvoBelief teaches LLM agents to test hypotheses about the world: 83.3% vs 0% for baselines
MagicBeanStalk lifts gpt-oss-120b from 61.2 to 69.8 points on math
CLEAT Teaches Qwen3-4B Agents to Ask for Clarification at the Right Time: Up to +13.2 pp
The CLEAT framework jointly trains a clarification controller and an agent policy. On Qwen3-4B it leads in all ten configurations and gains 13.2 pp on UserGym…
SpireBench Puts 20 LLMs Through a Full Run of Slay the Spire: Experts Still Ahead
The SpireBench benchmark runs 20 LLMs through full Slay the Spire at top difficulty: the best models match the average player but trail experts by a wide margin.
DBS chief: market is reassessing AI investment risk and will tighten financing
Refactoring C++ in OpenBMC: How an AI Agent Fought a 280-Line Function
CXMT claims breakthrough in 4F² memory cell architecture, prepares server DRAM for late 2026
Runpod shows how to run diffusion models on JAX in GPU pods
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…
EvoWeb Tests Coding Agents on 320 Sequential Requests Across 32 Web Apps
The new EvoWeb benchmark makes AI agents build out a single app step by step. The models' main weakness is integrating new features with code they have already…