AI news on Wednesday, June 24, 2026

SIGNIFICANCE

★ Top story · Hardware5d ago

OpenAI and Broadcom unveil LLM-optimized inference chip

OpenAI and Broadcom have unveiled Jalapeño, a custom semiconductor specifically engineered to handle the compute-heavy requirements of running large language models. By shifting from general-purpose hardware to specialized silicon, OpenAI aims to improve the performance and scalability of its AI services. The companies expect this hardware to reduce inference costs by 50 percent, marking a strategic move to lower operational expenses and decrease reliance on external chip suppliers for their core infrastructure.

OOpenAI Blog BBloomberg Technology TThe Decoder Read original ↗

Wednesday — June 24, 2026

OpenAI and Broadcom unveil LLM-optimized inference chip

Anthropic Accuses Alibaba of ‘Illicitly’ Accessing AI Models

Introducing computer use in Gemini 3.5 Flash

Figma bets on human judgment at Config 2026 while the AI powering its canvas belongs to someone else

Most major AI chatbots still lean left on political questions, even "anti-woke" models are no exception

China’s Zhipu AI Tops HK$1 Trillion on New Model Launch, Commercialization Hopes

AI Is Modeling Mega Storms With Surprising Accuracy

Repositioning retail for the AI era

OpenAI says ChatGPT Instant now better understands what users actually want

AI Is Designing Radio Chips That Humans Couldn’t Even Imagine

Facebook rolls out an AI companion app for creators

SiMa.ai and Mistral Solutions, a subsidiary of AXISCADES, Partner to Accelerate Autonomous Drone Intelligence

Dual-Branch Cross-Projection Debiasing through Diffusion-based Disentanglement

Tencent Tests DeepSeek-Powered AI Agent for WeChat Corporate App

NatureBench: Can Coding Agents Match the Published SOTA of Nature-Family Papers?

Universal Guideline-Driven Image Clustering via a Hybrid LLM Agent

A Benchmark for Hallucination Detection in VLMs for Gastrointestinal Endoscopy

CALIBER: Calibrating Confidence Before and After Reasoning in Language Models

Data Augmentation: A Fourier Analysis Perspective

MM-TRELLIS: Point-Cloud Guided Multi-Modal 3D Vehicle Generation in Autonomous Driving

Neuro-Symbolic Drive: Rule-Grounded Faithful Reasoning for Driving VLAs

Accelerating Multimodal Large Language Models with Prior-Corrected Token Reduction

ABACUS: Adapting Unified Foundation Model for Bridging Image Count Understanding and Generation

CompressKV: Semantic-Retrieval-Guided KV-Cache Compression for Resource-Efficient Long-Context LLM Inference

Agentic AI for Bilevel Long-Term Optimization of Policy-Driven Physical Layer Systems

RIFT-Bench: Dynamic Red-teaming For Agentic AI Systems

Sol Video Inference Engine: Agent-Native Full-Stack Acceleration Framework for Efficient Video Generation

OmniPath: A Multi-Modal Agentic Framework for Auditing Wheelchair Accessibility

MotifGen: Spatiotemporal interpolation of misaligned satellite images via multi-source generative modeling, in an application to tropical cyclones

Metis: Bridging Text and Code Memory for Self-Evolving Agents

When Helpfulness Overrides Causal Caution: Context-Dependent Suppression and Recovery in LLMs

GRACE: Gated Refinement for Accurate Causal Edge Discovery in High-Dimensional Time Series

A P\={a}ninian Foundation for Indic Language Processing

Bayesian control for coding agents

MedBench v5: A Dynamic, Process-Oriented, and Hallucination-Aware Benchmark for Clinical Multimodal Models

P-MTP: Efficient Document Parsing via Multi-Token Prediction with Progressive Depth Scaling

Natural Identifiers for Privacy and Data Audits in Large Language Models

An Introduction to Causal Reinforcement Learning

Faithful by Construction: Claim-Anchored Attribution for Multi-Document Summarization

Fast and Slow Variational Continual Learning

Gradium Launches stt-translate and s2s-translate, Real-Time Speech Translation Models Beating gpt-realtime-translate on Accuracy and Latency

Qwen-AgentWorld: Language World Models for General Agents

RoPE-Aware Bit Allocation for KV-Cache Quantization

ReMMD: Realistic Multilingual Multi-Image Agentic Verification for Multimodal Misinformation Detection

Ill-Posed by Design: Probing Evidence Use in VLMs

Governed Shared Memory for Multi-Agent LLM Systems

Trustworthy Image Authentication using Forensic Knowledge Graphs

TuringViT: Making SOTA Vision Transformers Accessible to All

One Year Later...The Harms Persist, But So Do We!

TIGER: Taming Identity, Geometry, and Generative Priors for High-Quality Face Video Restoration

Event-Grounded Question Answering over Long Audio via Structured Retrieval

How to Opt Out of Google Search’s New AI Data Training Feature

Warren Calls for AI Antitrust Scrutiny, Deal Review Revamp

Open-Vocabulary BEV Segmentation with 3D-Aware Geometric Constraints

NVIDIA and AWS Collaborate to Bring AI to Production at Scale

MGI: Member vs Generated Inference

Reinforcement Learning Towards Broadly and Persistently Beneficial Models

Pigeonholing: Bad prompts hurt models to collapse and make mistakes

Geometric Action Model for Robot Policy Learning

Medical AI Model Privacy Risks

Listening makes Vision Clear for VLMs

DriveStack-VLA: Render-Teacher Alignment for BEV-Based DeepStack Vision-Language-Action Model

Beyond Trajectory Imitation: Strategy-Guided Policy Optimization for LLM Reasoning

BehaviorBench: Benchmarking Foundation Models for Behavioral Science Tasks

ObsGraph: Hierarchical Observation Representation for Embodied Reasoning and Exploration

You Don't Need to Run Every Eval

Escaping the Self-Confirmation Trap: An Execute-Distill-Verify Paradigm for Agentic Experience Learning

CAVEWOMAN: How Large Language Models Behave Under Linguistic Input and Output Compression

LemonHarness Technical Report

SkyJEPA: Learning Long-Horizon World Models for Zero-Shot Sim-to-Real Control of Quadrotors

What Do Flow-Based Inverse Solvers Approximate? A Posterior-Transport View

The Geometry Behind Diffusion and Flow Matching: Gradient Flows and Geodesics in Wasserstein Space

SK Hynix Seeks $29 Billion With US Listing to Fund AI Boom

REDI-Match: Rotation-Equivariant Distillation for Efficient and Robust Dense Matching

OpenAI and Broadcom announce chip designed for LLM inference at scale

Survey finds 3 in 10 adults using social media, AI for health information and advice

AsyncOPD: How Stale Can On-Policy Distillation Be?

Learning to Trigger: Reinforcement Learning at the Large Hadron Collider

The Latent Bridge: A Continuous Slow-Fast Channel for Real-Time Game Agents