OpenAI Shuts Down Sora Video Tool
March 25, 2026
D.A.D. today covers 13 stories from 3 sources. What's New, What's Innovative, What's Controversial, What's in the Lab, and What's in Academe.
D.A.D. Joke of the Day: My AI said it needed more context to help me. I gave it 47 emails, 12 attachments, and my calendar. Now it needs a mental health day.
What's New
AI developments from the last 24 hours
OpenAI Shuts Down Sora Video Tool Months After Launch
OpenAI is shutting down Sora, its AI video generation tool, just months after launching it publicly and announcing a Disney partnership. The company hasn't provided detailed reasoning, but user speculation points to high operational costs, restrictive content policies, and intensifying competition from Chinese alternatives like Seedance 2.0, which some users say has surpassed Sora's capabilities. Community reaction included questions about why Sora was spun out as a separate app rather than integrated into ChatGPT, and speculation that OpenAI is retooling to focus on a narrower target: business audiences, given increased competition from Anthropic.
Why it matters: The rapid rise-and-fall signals that even well-resourced AI labs are struggling to make video generation commercially viable—and that Chinese competitors may be setting the pace in this space.
Discuss on Hacker News · Source: twitter.com
Linux Update Brings Major Speed Gains for Windows Games
Wine 11 has shipped with NTSYNC, a long-awaited feature that changes how Linux handles synchronization for multi-threaded Windows games. The previous workarounds had practical limits—one hit file descriptor caps, the other required kernel patches that never made it into mainline Linux. The developers say performance improvements range from "noticeable to absurd" for games that benefit, though not every title will see dramatic gains. The release also completes a major architecture overhaul and improves support for Wayland, the modern Linux display system.
Why it matters: This is Linux gaming infrastructure—if you or your team game on Linux or deploy Windows apps via Wine, expect smoother performance on demanding titles without needing custom kernel patches.
Discuss on Hacker News · Source: xda-developers.com
Apple Launches All-in-One Business Platform for Companies Without IT Staff
Apple announced Apple Business, an all-in-one platform launching April 14, 2026, that bundles mobile device management, business email with custom domains, and customer outreach tools. The service targets companies without dedicated IT staff, offering zero-touch device deployment and integration with Google Workspace and Microsoft Entra ID. A new local advertising option in Apple Maps arrives this summer for U.S. and Canadian businesses. The platform will be available in over 200 countries.
Why it matters: Apple is positioning itself as a direct competitor to Microsoft 365 and Google Workspace for small and mid-sized businesses, bundling the MDM capabilities that previously required third-party tools—potentially simplifying the stack for Apple-first organizations while locking them deeper into the ecosystem.
Discuss on Hacker News · Source: apple.com
What's Innovative
Clever new use cases for AI
Gemini Can Now Search Raw Video With Plain English Queries
A developer built a command-line tool demonstrating Gemini Embedding 2's ability to embed raw video directly into searchable vector space—no transcription or frame captioning required. You can index footage and search it with natural language queries like 'person carrying a box,' with results trimmed to matching clips. Indexing runs about $2.50 per hour of footage, with smart frame-skipping to reduce costs on static security camera content. Early interest on Hacker News focused on home monitoring use cases, though questions about handling low-confidence matches went unanswered.
Why it matters: If the capability holds up, this could dramatically simplify video search workflows for security footage, content libraries, or compliance archives—previously you needed separate transcription, object detection, or manual tagging to make video searchable.
Discuss on Hacker News · Source: github.com
ProofShot Lets AI Coding Agents Capture Screenshots of What They Build
A developer released ProofShot, an open-source command-line tool that lets AI coding agents like Claude Code, Cursor, or Codex open a browser, interact with pages, and capture video, screenshots, and console logs into a single HTML file for human review. The tool works via shell commands, making it agent-agnostic. Community reaction on Hacker News was skeptical—users questioned differentiation from existing tools like playwright-cli, noted that exe.dev already offers similar functionality natively, and debated whether capturing screenshots without built-in analysis constitutes genuine visual verification.
Why it matters: This is developer plumbing for AI-assisted coding workflows—useful if you're building interfaces with AI agents and need a streamlined way to generate visual proof of what they created, though the crowded space of similar tools suggests you should evaluate alternatives before committing.
Discuss on Hacker News · Source: github.com
Flighty Adds Real-Time Airport Delay Maps for Travelers
Flighty, the flight tracking app, has added an airports feature showing real-time cancellation and delay rates on an interactive map. Users can see at-a-glance percentages for airports, with flags like 'High Cancellations' when rates hit thresholds (33% in examples shown). Early user reaction has been enthusiastic, with some calling it 'one of the best apps ever,' though testers noted minor display bugs with airport ordering.
Why it matters: For frequent travelers, this gives a quick visual check of airport conditions before heading out—useful for deciding whether to leave early or expect disruptions.
Discuss on Hacker News · Source: flighty.com
What's in the Lab
New announcements from major AI labs
OpenAI Releases Free Tool to Help Apps Protect Teen Users
OpenAI released gpt-oss-safeguard, an open-source tool providing prompt-based safety policies designed to help developers moderate age-specific risks for teenage users. The release includes guidelines for handling sensitive topics when AI systems detect they're interacting with minors. No technical benchmarks or effectiveness data accompanied the announcement.
Why it matters: As regulators globally scrutinize how AI platforms handle minors—and as more consumer apps integrate AI features—this gives developers a starting framework, though its real-world effectiveness remains unproven.
OpenAI Foundation Commits $1 Billion to Health and Economic Programs
OpenAI announced its foundation arm will invest at least $1 billion in philanthropic initiatives across four areas: curing diseases, expanding economic opportunity, building AI resilience, and community programs. The commitment comes as OpenAI restructures its corporate organization and faces ongoing questions about balancing its nonprofit origins with commercial growth. No timeline or specific project details were provided.
Why it matters: The billion-dollar pledge signals OpenAI's effort to maintain its public-benefit identity as it scales commercially—a positioning move as regulatory and public scrutiny of AI labs intensifies.
ChatGPT Adds Visual Shopping and Product Comparison Features
OpenAI launched enhanced shopping features in ChatGPT, built on what it calls the Agentic Commerce Protocol. Users can now browse products, compare items side-by-side, and interact with integrated merchants directly within the chat interface. The company says the protocol enables richer product discovery and a more visual shopping experience. No details were provided on which merchants have integrated or how the protocol works technically.
Why it matters: OpenAI is positioning ChatGPT as a commerce platform, not just a chatbot—a move that could eventually compete with Google Shopping and Amazon's product search if merchant adoption scales.
What's in Academe
New papers on AI and its effects from researchers
Touch-Enabled Robots Handle Delicate Tasks 80% Better Than Vision-Only Systems
Researchers developed VTAM (Video-Tactile Action Model), a framework that adds touch sensing to robot vision systems for tasks requiring delicate physical contact. The approach addresses a key limitation of vision-only systems: they can't tell how much force a robot is applying. In tests, VTAM achieved 90% success on contact-sensitive manipulation tasks. On particularly delicate work—like picking up potato chips without crushing them—it outperformed a vision-only baseline by 80%.
Why it matters: This is robotics research, not a product you'll encounter soon—but it signals progress toward robots that can handle fragile objects in warehouses, manufacturing, and eventually service settings where 'grab it without breaking it' is the baseline requirement.
Training Technique Cuts AI Reasoning Overhead by Half
Researchers developed SortedRL, a scheduling technique that speeds up reinforcement learning training for AI models by more efficiently handling the long chains of reasoning these systems generate. The method cuts wasted compute time (called 'bubble ratio') by over 50% while actually improving model performance by 4-18% on math and logic benchmarks. The gains matter because generating extended reasoning chains—which can run 16,000 tokens—currently consumes up to 70% of training time.
Why it matters: This is infrastructure research, but it signals that training reasoning-capable models could become significantly cheaper and faster—potentially accelerating how quickly labs can develop and refine the chain-of-thought capabilities behind tools like ChatGPT's reasoning mode.
Researchers Find Way to Reuse Training Data for Reasoning Models
Academic researchers propose ReVal, a training approach for AI reasoning models that reuses past training data instead of generating fresh examples for each learning cycle. In tests on DeepSeek's 1.5B-parameter model, ReVal showed 2.7% improvement on a math benchmark and 4.5% on a science reasoning test compared to current standard methods, while converging faster. The technique addresses a known inefficiency in how today's reasoning models are trained—they typically discard data after one use.
Why it matters: This is training infrastructure research; if validated at scale, it could reduce the compute costs of building reasoning-capable AI models, potentially lowering barriers for smaller labs.
Users Judge Earthquake Alerts by Speed, Not Technical Accuracy
Researchers used LLMs to analyze 500+ social media posts about Android's earthquake alert system during Turkey's April 2025 Marmara earthquake—the region's largest in 25 years. The key finding: users judge alert accuracy primarily by timeliness, not technical precision. Even when the system correctly predicted shaking and provided over a minute of warning, users who received alerts after feeling tremors rated the system as inaccurate. The study suggests engineering teams and end users define 'accurate' very differently.
Why it matters: For any organization deploying AI-powered alerting or notification systems, this research suggests that perceived reliability hinges on timing—a technically correct alert that arrives late may damage user trust as much as a wrong one.
AI Framework Aims to Understand Entire Cities in 3D
Researchers built 3DCity-LLM, a framework that lets AI models understand and reason about entire cities in 3D—not just individual buildings or streets, but urban environments at scale. The accompanying dataset contains 1.2 million training samples spanning tasks from analyzing specific objects to planning across complex scenes. The team claims it significantly outperforms existing methods on spatial reasoning benchmarks, though the paper doesn't provide specific numbers in its abstract.
Why it matters: This is research-stage work, but city-scale 3D understanding could eventually reshape urban planning tools, autonomous vehicle navigation, and real estate analytics—fields where professionals currently stitch together 2D maps and fragmented data sources.
What's On The Pod
Some new podcast episodes
AI in Business — From AI Experiments to Enterprise Value Driving Real Business ROI - with Dan Diasio of EY
How I AI — How Microsoft's AI VP automates everything with Warp | Marco Casalaina