Research

Everything in this section in one place — newest stories first.

LatestLLMs

Apple Clears China's AI Hurdle by Partnering with Alibaba's Qwen

China's Cyberspace Administration approved Apple Intelligence for launch after Apple agreed to integrate Alibaba's Qwen model across iOS, iPadOS, macOS, and visionOS.

LLMsJuly 15, 2026

New Benchmark Exposes a Blind Spot: LLMs Struggle to Tutor Like Teachers, Not Just Talk Like Them

CSTutorBench, a new benchmark from H. Chad Lane and Bryson Kageler, tests 11 LLMs (4B to 120B parameters) as CS tutors inside the VEX VR robotics environment using 17 scenario-based questions.

LLMsJuly 15, 2026

Grok 4.5's Real Edge Isn't the Benchmark Score, It's the Bill

Grok 4.5 prices in at $2/$6 per million input/output tokens, far undercutting Opus 4.8 ($5/$25), Fable 5 ($10/$50), and GPT-5.5/5.6 ($5/$30)

ResearchJuly 15, 2026

MIT Taps AI-Physics Pioneer Jesse Thaler to Lead Its Nuclear Science Laboratory

Jesse Thaler becomes director of MIT's Laboratory for Nuclear Science on August 1, 2026, succeeding Bolek Wyslouch after his decade-long tenure.

AI AgentsJuly 15, 2026

Three New Benchmarks Put LLMs to Work on Industrial CAD and Systems Engineering

ArtisanCAD combines a new CAD-IR intermediate representation with a CATIA-MCP backend, cutting Chamfer Distance from 14.83 to 9.88 on the Text2CAD benchmark for industrial part generation.

AI AgentsJuly 15, 2026

Physics-Inspired Attribution Method Ditches Causal Graphs for Explaining IoT Failures

A new arXiv paper (2607.05563) proposes an energy-based, undirected alternative to causal graphs for explaining behavior in cyber-physical IoT systems.

ResearchJuly 15, 2026

Why Being Right Doesn't Always Pay: New Research Fixes a Broken Link in Prediction Markets

New research formalizes why accurate forecasters often lose money on order-book-based prediction markets like Kalshi and Polymarket.

ResearchJuly 15, 2026

Can LLMs Fake Consumer Psychology? New Study Tests AI Against Human Focus Groups

A new arXiv paper (2607.05761, posted July 7, 2026) tests whether LLMs can replicate human responses in projective consumer research techniques.

ResearchJuly 15, 2026

A New Loss Function Aims to Sharpen AI's Ear for Depression Detection

A new arXiv paper by Manning Gao introduces a fine-grained multimodal framework for detecting depression from combined audio and visual data.

LLMsJuly 15, 2026

Four New Benchmarks Expose How Little We Actually Know About What LLMs Are "Thinking"

CCBENCH-Health finds top LLMs give culturally appropriate healthcare advice only 20-30% of the time, dropping to 8.8% for Afghan personas

RoboticsJuly 15, 2026

Robotics Hits an Inflection Point: IPOs, Billion-Dollar Bets, and Battlefield-Tested Autonomy All Converge

Agility Robotics is going public via a $2.5B SPAC merger with Churchill Capital XI, raising over $620M, while Unitree cleared a Shanghai IPO at a ~$5.9B valuation.

ResearchJuly 15, 2026

One-Step Diffusion Model OSOR Cuts Object Removal Times by Up to 30x

OSOR (One-Step Object Removal) is a diffusion model that performs object removal in a single step, running 4x to 30x faster than multi-step diffusion baselines while matching or beating their quality.

ResearchJuly 15, 2026

A Compiler for Trust: New Framework Treats "Content Slop" as a Syntax Error, Not a Style Problem

CANONIC is a new governance framework that verifies digital artifacts using compiler-style structural checks instead of prose-based content review

ResearchJuly 15, 2026

The 'Granularity Paradox': Why Finer Time-Series Data Can Make Forecasts Worse

A new arXiv paper identifies a 'Granularity Paradox': finer time intervals (e.g., daily vs. monthly) improve in-sample metrics but often worsen real-world forecast accuracy due to recursive error compounding.

EthicsJuly 15, 2026

Researchers Confront AI's CSAM Problem With a Position Paper and a No-Image Audit Method

A position paper spotlighted at ICML 2026 identifies 15 unsolved problems in preventing AI-facilitated child sexual abuse material across the model development lifecycle.

ResearchJuly 15, 2026

New Graph Clustering Framework SCISE Tackles the "Structural Isolation" Problem in Mini-Batch Training

SCISE is a new unsupervised graph clustering framework accepted to VLDB 2026, aimed at fixing 'structural isolation' in mini-batch Graph Contrastive Learning.

EthicsJuly 15, 2026

ChatGPT Logs Reveal Behavioral Fingerprints of Depression, But Fall Short as a Screening Tool

A study of 187,093 ChatGPT conversations from 766 users linked PHQ-8 depression scores to distinct usage patterns, including late-night sessions and higher rates of self-disclosure.

LLMsJuly 15, 2026

A New Research Blueprint Asks: Can AI Actually Understand India's Languages, or Just Process Them?

A new arXiv paper reframes Indic AI development as a cultural heritage preservation challenge rather than a pure scaling problem.

All articles