โ˜•๏ธ OpenAI's Sora burned $1 million daily before shutdown

OpenAI's costly Sora, Apple's Siri store, Microsoft ads, and more

โ˜•๏ธ OpenAI's Sora burned $1 million daily before shutdown

Hi there, this is your daily โ˜•๏ธ Techpresso.


In today's Techpresso:

๐Ÿ’ธ OpenAI's Sora burned $1 million daily before shutdown

๐ŸŽ Apple may launch a dedicated Siri AI app store

๐Ÿค‘ Microsoft Copilot now shows ads in pull requests

๐Ÿ“‰ ChatGPT app store struggles six months post-launch

๐Ÿ’Š Eli Lilly bets $2.75 billion on AI drug discovery

Plus: ๐ŸŽ 14 other news you might like, ๐Ÿงฐ 6 tools, and ๐Ÿ“š 3 papers.

๐Ÿ’ธ OpenAI's Sora burned $1 million daily before shutdown LINK

  • OpenAI's video-generation tool Sora was losing roughly a million dollars per day before the company shut it down last week, just six months after its public launch.
  • Sora's user count peaked at around one million but then dropped below 500,000, while video generation kept consuming expensive AI chips at a rate OpenAI could not justify.
  • Disney had committed $1 billion to a partnership with OpenAI around Sora but learned about the shutdown less than an hour before the public announcement, killing the deal.
  • ๐ŸŽ Apple may launch a dedicated Siri AI app store LINK

  • Apple is reportedly planning to let users install third-party AI chatbots inside Siri through a new feature called "Extensions," with a dedicated section in the App Store acting as an AI marketplace.
  • Bloomberg's Mark Gurman reports that the changes are coming in iOS 27, which should get its first developer beta in June around Apple's Worldwide Developers Conference starting June 8.
  • Apple already integrated OpenAI's ChatGPT into Siri, but Extensions will offer a wider choice of AI agents that users can browse and install from the new App Store section.
  • ๐Ÿค‘ Microsoft Copilot now shows ads in pull requests LINK

  • Microsoft's Copilot AI tool has started inserting promotional messages into pull requests on GitHub, advertising a Raycast extension for the Copilot coding agent on macOS and Windows machines.
  • A search on GitHub shows the exact same promotional text appearing in over 11,000 different pull requests across thousands of repos, and even merge requests on GitLab contain the injection.
  • Hidden HTML comments labeled "START COPILOT CODING AGENT TIPS" in the raw markdown suggest Microsoft is inserting these ads, likely to promote its own developer ecosystem or partner integrations.
  • ๐Ÿ“‰ ChatGPT app store struggles six months post-launch LINK

  • OpenAI's push to turn ChatGPT into an app platform with over 300 integrations from companies like Booking and StubHub is off to a sluggish start six months after launch.
  • Partner companies are hesitant to hand off customer relationships and payments to OpenAI, so most apps force users to leave ChatGPT to complete purchases or even view basic details.
  • Developers have complained about a tedious app-approval process, buggy development tools, and a lack of usage data, leaving them "running quite blind" on whether their apps are working properly.
  • ๐Ÿ’Š Eli Lilly bets $2.75 billion on AI drug discovery LINK

  • Eli Lilly announced a $2.75 billion deal with Hong Kong-based Insilico Medicine to use artificial intelligence for drug discovery and development, giving Lilly exclusive license to manufacture and market the resulting therapies.
  • Insilico CEO Alex Zhavoronkov told CNBC his company has used AI to develop at least 28 drugs, with close to half already at a clinical stage of testing.
  • Eli Lilly has also committed $1 billion over five years with Nvidia to finance talent, infrastructure and computing needed to address bottlenecks in AI-based drug discovery.
  • Other news you might like

    • Chinaโ€™s DeepSeek suffers rare outage lasting several hoursLINK
    • Mistral AI raises $830M in debt to set up a data center near ParisLINK
    • Bluesky built a new AI tool that wants to free you from social algorithmsLINK
    • Big tech was embracing clean energy and turning a corner on climate change. Then AI data centers arrivedLINK
    • Agibot Shipped A Staggering 5,000 Humanoid Robots In The Last 3 MonthsLINK
    • 'Weekend work increased more than 40%': How AI is quietly expanding the work week as Saturday starts shift to 7:11 amLINK
    • Exclusive: Android could soon get its own AirDrop-style โ€˜tap to shareโ€™ featureLINK
    • GPT drafts, Claude critiques: Microsoft blends rival AI models in new Copilot upgradeLINK
    • Apple Reportedly Planning to Launch iMac With OLED DisplayLINK
    • An AI Agent Was Banned From Creating Wikipedia Articles, Then Wrote Angry Blogs About Being BannedLINK
    • EVs were meant to bypass oil. Now theyโ€™re stuck at the Strait of HormuzLINK

    ๐Ÿงฐ Trending tools

    PopTask: a macOS menu bar task manager that uses AI to parse natural language input and automatically schedule tasks across seven languages.LINK

    Goals: breaks down objectives into daily AI-generated action steps with streak tracking, helping you maintain consistent progress without managing complex task lists.LINK

    Invoke: a desktop IDE combining visual canvas planning with multi-model AI code generation to streamline feature design and implementation workflows.LINK

    Blood Sugar Journal: tracks glucose and insulin data with AI-powered analysis to generate actionable reports, featuring iCloud sync for iOS users managing diabetes.LINK

    Bluor AI: generates responsive, on-brand email designs from text prompts in under 60 seconds, eliminating manual coding and design work for marketers.LINK

    Crazierl: an experimental operating system built around the BEAM VM that boots on x86 hardware and supports Erlang distributed clustering.LINK

    ๐Ÿ“š Trending papers & reports

    Mathematical proof methods remain fundamentally different from AI pattern matching because they require understanding why something is true, not just recognizing that it appears true from examples.LINK

    LLM essay grading correlates poorly with human teachers, agreeing only 44% of the time on scores compared to 65% agreement between two human graders in standardized tests.LINK

    Large language models regress toward average responses when trained on diverse data because they learn to predict the mean of their training distribution, explaining why they often produce generic outputs.LINK


    Want to get the latest news differently? Find us on:

    twitter instagram spotify apple-podcasts


    See you tomorrow for a new dose of โ˜•๏ธ Techpresso!

    More from the archive