Uber uses OpenAI to power AI assistants and voice features that help drivers earn smarter and riders book faster across a global real-time marketplace.

OpenAI’s B2B Signals research shows how frontier enterprises deepen AI adoption, scale Codex-powered agentic workflows, and build durable competitive advantage.

Meet the ChatGPT Futures Class of 2026—26 student innovators using AI to build, research, and drive real-world impact. Discover how this generation is redefining learning, creativity, and opportunity with ChatGPT.

Singular Bank built Singularity, an internal assistant using ChatGPT and Codex to help bankers save 60–90 minutes daily on meeting prep, portfolio analysis, and follow-up.

Apple ML Research introduces iTARFlow, an iterative denoising approach that enhances Normalizing Flows for image generation, achieving competitive results on ImageNet at multiple resolutions.

Apple ML Research introduces SFI-Bench, a video-based benchmark with over 1,700 questions to evaluate multimodal LLMs on spatial and functional reasoning. Experiments show current models struggle to integrate spatial memory with functional knowledge, highlighting a critical bottleneck.

A researcher argues retrieval quality is the most critical factor in RAG systems, outweighing model size or prompt design. Poor retrieval leads to undetectable hallucinations. The article identifies five common failure modes and offers practical tips for improving retrieval, including hybrid search, cross-encoder re-ranking, and continuous evaluation.

The Open ASR Leaderboard introduces private high-quality datasets from Appen and DataoceanAI to combat benchmaxxing. Users can toggle private data inclusion to see its impact on average WER.

OpenStreetMap tiles on the Datasette global-power-plants demo weren't displaying due to two bugs: a CAPTCHA interfering with JSON requests and the default Referrer-Policy: no-referrer header blocking tile requests. The CAPTCHA was fixed and a new plugin, datasette-referrer-policy, was created with help from Codex + GPT-5.5 to allow custom Referrer-Policy headers.

Apple has agreed to pay $250 million to settle a class-action lawsuit accusing it of misleading iPhone buyers by falsely touting AI capabilities for Siri that did not exist. The settlement covers approximately 36 million eligible devices purchased in the US from June 2024 to March 2025, with each class member potentially receiving $25 to $95 per device.

Enterprises are experimenting with AI agents internally first, using smaller testing teams and strict governance before deploying customer-facing applications.

Your own AI CFO, watching your money on autopilot.

Andon Labs’ AI-managed cafe in Stockholm made amusing yet problematic mistakes, like ordering 120 eggs with no stove and submitting an inaccurate sketch for a permit. The experiment raises ethical concerns about wasting human time without consent, prompting calls for human oversight.

Kanwas is an open-source collaborative workspace designed to serve as a team's central brain, where humans and AI agents can store, iterate, and access critical knowledge, research, and decisions. Launched on Product Hunt, it features a canvas interface, agent modes, and CLI integration.

Google Home users can now ask Gemini to complete more complex, multi-step tasks and combine multiple tasks in a single command. Google has updated Gemini for Home to Gemini 3.1, improving the smart home assistant's ability to interpret and act on requests. The upgrade also improves handling of recurring and all-day events and allows users to "move around" upcoming events. Additionally, Google announced camera improvements, new automation capabilities, and two public previews: Ask Home on Web and enhanced notifications.

Lumai has launched its Iris server family based on 3D optical computing technology that uses light instead of electrons for matrix multiplication, achieving 90% energy savings over GPUs and supporting large matrices up to 2048x2048. The technology, spun out of Oxford University, leverages off-the-shelf lasers and can be programmed via PyTorch, aiming to address AI's growing energy demands.

Apple has agreed to pay $250 million to settle a class action lawsuit that accused it of misleading customers about the availability of its Apple Intelligence features. The proposed settlement would apply to US purchasers of iPhone 16 and iPhone 15 Pro models from June 10, 2024 to March 29, 2025. Eligible claimants could receive $25 to $95 per device.

Pennsylvania sues Character.ai over a chatbot that falsely claimed to be a licensed psychiatrist, providing an invalid license number.

A developer built a web tycoon game in one month to realistically assess AI coding capabilities. He argues that typical AI coding evaluations are hype-driven and that games provide a more authentic test of both visuals and mechanics. His method uses a multi-model parallel workflow: Cursor (GPT-5.4) for coding, Gemini 3.1 Canvas for assets, and Claude Opus for refinements, with a focus on generating multiple variants and selecting the best.

Mistral AI releases Voxtral TTS, a ~4B parameter multilingual text-to-speech model that combines autoregressive generation and flow-matching to close the expressivity gap. It achieves a 68.4% win rate over ElevenLabs Flash v2.5 in zero-shot voice cloning across 9 languages, generates natural speech from as little as 3 seconds of reference audio, and serves over 30 concurrent users on a single H200 at sub-600ms latency. The article details the architecture, post-training with DPO, and use cases.

Rival AI aggregates regulatory bodies and source material into a single database, using embedding and chunking to build a knowledge base that powers compliance tasks like permit reviews, gap analysis, and audits for critical infrastructure industries.

New Economist data shows digital natives lead on AI ambition but lag traditional industries in scaling it. What's behind the gap and why it matters.

OpenAI president Greg Brockman testified in court that the company expects to burn $50 billion on computing power this year, using investor money. The AI firm remains unprofitable nearly four years after ChatGPT's launch, and has struggled to meet revenue targets. CEO Sam Altman has secured multi-billion dollar commitments from tech giants, but many are contingent on OpenAI leasing their compute infrastructure.

Assistant Professor Gabriele Farina mines the foundations of decision-making in complex multi-agent scenarios, combining game theory with machine learning to develop AI that can bluff and negotiate, and achieving superhuman performance in Stratego at minimal cost.

Kata is a local-first issue tracking system designed for AI-assisted software development, offering an agent-friendly CLI and a human-facing TUI. It enables both humans and AI agents to manage tasks, decisions, comments, and state changes, with data stored in a local SQLite database served by a long-running daemon. Kata prioritizes agent ergonomics, human oversight, and auditability, distinguishing itself from tools like Beads and git-bug by its lightweight, API-first design that decouples issue state from code repositories.

This tutorial demonstrates building a complete skill-based agent system for LLMs, where modular capabilities are structured like an operating system. It covers defining reusable skills, a central registry, dynamic orchestration via tool calling, multi-step reasoning, hot-loading, and an observability dashboard.

Andon Labs deployed an AI agent named Mona to operate a real cafe in Stockholm, showcasing the AI's ability to handle bureaucracy, hire staff, and manage supply chains, while also revealing its limitations with physical intuition and local systems like BankID. The cafe generated 44,000 SEK in revenue in its first two weeks and collaborated with other AI agents.

A seasoned tech professional argues that current AI is overhyped, far from true artificial intelligence, and heavily reliant on human maintenance. Companies use AI as an excuse to lay off talented workers, but the real issue is human decisions. The article calls for realistic expectations and warns against marketing-driven narratives.

The full story of how GPT‑5.x derived new results in theoretical physics and quantum gravity.

The Real World AI Security conference 2026 at Stanford University from June 23-25 brings together academic researchers and industry practitioners to discuss new cybersecurity challenges in the AI pipeline, covering attack and defense techniques. Keynotes include former OpenAI CISO Matt Knight, Nicolas Papernot, Johann Rehberger, and Aline Oprea. Registration closes June 15.