All articles

The complete WAKIB archive: every published story, across every section, newest first.

LatestResearch

What Makes a Benchmark Actually Good? New arXiv Paper Lays Out the Criteria

A new arXiv paper, 'Good Benchmarks' (2607.12217), submitted by Ivan Bercovich on July 13, 2026, outlines five criteria for evaluating AI benchmark quality.

LLMsJuly 22, 2026

New Benchmark Exposes a Blind Spot in AI Agents: Remembering to Do Things Later

PM-Bench, a new text-based benchmark inspired by cognitive science's Virtual Week paradigm, tests whether LLM agents can remember and execute delayed intentions over a simulated seven-day period.

ResearchJuly 22, 2026

A Penalty Term Could Fix Federated Learning's Blind Spot for Grid Safety

A new arXiv preprint (2607.12763) proposes a penalty-based aggregation rule for federated reinforcement learning that accounts for system-level safety constraints, unlike standard FedAvg.

RoboticsJuly 22, 2026

Hierarchical Planning Boosts Long-Horizon Robot Control by Up to 14.7 Points

Hi-LeWM adds a high-level planner over frozen LeWorldModel latents to handle long-horizon goal-conditioned control.

ResearchJuly 22, 2026

More Data, More Confusion: New Study Shows Bayesian Causal Discovery Can Break Down as Sample Size Grows

A new arXiv paper (submitted July 10, 2026) derives an exact critical correlation threshold at which Bayesian causal discovery mistakes latent confounding for a direct causal edge.

RoboticsJuly 22, 2026

Researchers Detail Unified Guidance System for Fixed-Wing UAVs to Track and Engage Targets Autonomously

A new arXiv paper (2607.12801), submitted July 14, 2026, presents a unified control framework for fixed-wing UAVs handling detection, tracking, and terminal engagement in one system.

AI InfrastructureJuly 22, 2026

Morgan Stanley Turns AI Data Centers Into Wall Street's Newest Bond Machine

Morgan Stanley's capital-markets fees rose to $2.3 billion in H1, up from $1.4 billion a year earlier, pushing it past Goldman Sachs into second place behind JPMorgan.

LLMsJuly 22, 2026

When LLMs Say "I'm Sure," Do They Actually Know? A New Study Measures the Gap

Researchers applied Signal Detection Theory to separate LLM answer accuracy from the reliability of the model's own confidence scores across 224,000 QA trials.

Healthcare AIJuly 22, 2026

Elevate Capital Buys Egypt's Nile Radiology in $19.5M Deal, Eyes AI-Driven Diagnostics and EGX Listing

Elevate Capital and the Moharram family signed a definitive agreement to acquire 100% of Nile Radiology and Labs for approximately USD 19.5 million (EGP 1 billion)

ResearchJuly 22, 2026

When Audio AI Hears "No Dog Barking" as "Dog Barking"

NegEval-Audio is a new benchmark exposing that audio-language embedding models like CLAP fail to understand negation, mapping 'dog barking' and 'no dog barking' to nearly identical embeddings.

LLMsJuly 22, 2026

Ask an LLM to Free-Associate and It Thinks Smaller Than You Do

A new study compares semantic word-association patterns between 82 human participants and three LLMs (GPT-4o, Gemini-2.5-Pro, Claude-Sonnet-4.5)

ResearchJuly 22, 2026

New Hybrid Model Pairs Physics Simulation with AI to Predict Battery Aging

BattVAE-GP is a new hybrid framework combining physics-based battery simulation with probabilistic machine learning to model lithium-ion degradation.

AI AgentsJuly 22, 2026

OSWorld 2.0 Benchmark Exposes How Far Computer-Use AI Still Has to Go

OSWorld 2.0 introduces 108 long-horizon computer-use tasks that take humans a median of 1.6 hours, far exceeding the original benchmark's difficulty.

AI AgentsJuly 22, 2026

MBZUAI's QRAFT Puts AI Fact-Checkers to the Test, With Human Editors as the Judge

MBZUAI researchers built QRAFT, a three-agent AI framework (planner, writer, editor) designed to draft fact-checking articles like human journalists do.

LLMsJuly 22, 2026

What Makes an AI Model "Know" You? A New Study Says It's Not Your Résumé

A new arXiv paper introduces NameRank, a method for measuring how LLMs recognize people and entities.

AI AgentsJuly 22, 2026

Vercel Buys DevOps Agent Startup Stakpak, Its Third Acquisition in Weeks

Vercel has acquired Stakpak, an Egyptian-founded autonomous DevOps AI agent startup, with financial terms undisclosed.

ResearchJuly 22, 2026

A More Flexible Recipe for Semi-Supervised Learning: New Risk Rewriting Framework Beats Binary Limits

A new arXiv paper generalizes PNU (positive-negative-unlabeled) learning, a distribution-free semi-supervised technique previously limited to binary classification, extending it to multiclass problems.

LLMsJuly 22, 2026

A New Distance-Based Framework Tackles One of Machine Learning's Trickiest Uncertainty Problems

A new arXiv paper by Xabier Gonzalez-Garcia introduces a distance-based method to measure total, aleatoric, and epistemic uncertainty in multiclass classification using credal sets.

All articles