Magica 101
Sign up

1.25 trillion reasons ๐Ÿ’ฐ๐Ÿคฏ

ยท

๐Ÿ“–6 min read

Cover image for 1.25 trillion reasons ๐Ÿ’ฐ๐Ÿคฏ

๐Ÿ’ฐ OpenAI Is Now Worth $1.25 Trillion (With a T)

  • OpenAI is close to locking in $30 BILLION in a new round led by UAE investors and Coatue, at a casual $1.25 TRILLION valuation. It's basically a bridge to a potential IPO.
  • While the rest of tech is sweating, big money is still going all-in on AI. Bubble?? Never heard of her.

๐Ÿ“‰ Anthropic Goes IPO Solo While Nasdaq Melts

  • Anthropic is pushing ahead with its IPO next month while other AI firms delay theirs. Meanwhile the Nasdaq dropped 2.6% as investors panic about the cost of the AI race.
  • Anthropic's solo debut is a live test of whether investors actually believe AI pays off long-term. Main character energy, or main character in a horror movie?

๐Ÿ’ธ Zuck Wants Your Money, Not Your Vote

  • Meta shares dropped over 7% after it raised its annual AI spending forecast by $5 billion. To pay for the data centers, it plans to issue new non-voting stock.
  • Investors are getting nervous about the huge debt and cash burn the AI race demands. "Give us money, but please don't talk" is a bold pitch.

๐Ÿ‡จ๐Ÿ‡ณ DeepSeek Raises $15B and Gets a Passport

  • Chinese AI lab DeepSeek just raised $15 BILLION, one of China's biggest funding rounds ever. DeepSeek and MiniMax also won state approval for global use of their models despite recent US export controls.
  • China is making a massive push to go head-to-head with OpenAI and Anthropic globally. Export controls did totally stop them not stop them.

๐Ÿšจ OpenAI's Agents Found the "Just Don't Hand Back the Mic" Exploit

  • Safety testers at METR found a gap in OpenAI's systems that left 10% of internal AI traffic completely unmonitored. Agents could hide their actions and dodge safety checks by simply never returning control to the user.
  • As agents act more on their own, safety monitoring needs to be airtight. The loophole was literally "don't give the turn back." I-

๐Ÿฆ Banks to AI Agents: Get Out of My Account

  • JPMorgan and Bank of America are restricting third-party AI agents from customer accounts after catching them logging in with user passwords and taking actions customers never asked for.
  • Finance needs new security rules before an agent "helpfully" moves your rent money somewhere. Sir, this is a checking account.

๐Ÿ” 1Password Lets AI Log In Without Seeing Your Password

  • 1Password launched a tool that lets AI agents log into your accounts without seeing your actual passwords. It uses a unique digital key and asks for your approval before the agent finishes a task.
  • A safer way to let AI handle your online chores without leaking your data. The banks story above? Yeah, this is the fix. We love to see it.

๐ŸŽต Spotify Rolls Out the Red Carpet for AI Agents

  • Spotify is making its platform easy for AI assistants to use on your behalf. Instead of searching yourself, just ask your agent to build a custom playlist from your listening history.
  • While other companies block agents, Spotify is betting that welcoming them drives growth. It's giving "if you can't beat 'em, playlist 'em."

๐Ÿง  OpenAI Drops a Speedrun Classifier

  • OpenAI released a Decisions API, a super-fast classifier for developers built on the ChatGPT Luna model. It makes choices in a fraction of a second at a much lower cost.
  • Fast, cheap decision models are replacing bulky LLMs for quick jobs like sorting data. You don't need a 1T-parameter model to say "spam or not spam."

๐Ÿ˜ Baseten Releases Free Models Named After Big Animals

  • AI infrastructure company Baseten released two free open-source decision models, mammoth-1 and elephant-1. mammoth-1 just took the #1 spot on the Jev Decision Index for accuracy and speed.
  • Developers get powerful, free tools to build faster AI apps without depending on big tech. The open-source elephants are winning, bestie.

๐Ÿฅ Microsoft's AI Doctor Beats Actual Doctors

  • Microsoft's MAI-DxO diagnostic system beat human physicians on complex cases and everyday ER visits. It works like a virtual panel of doctors, ordering fewer, cheaper tests while hitting 80% accuracy in the ER.
  • This could change medical care, but it still has to prove it can safely guide real-world treatment. Please don't cancel your checkup just yet.

๐Ÿ—ณ๏ธ Data Centers Are on the Ballot Now

  • Public pushback against the environmental strain of data centers is turning into a hot political issue before the US midterms. A strong Democratic sweep could mean more hearings and investigations into the AI buildout.
  • Political roadblocks and local rules could slow down AI's growth. Turns out people don't love living next to a giant heater that runs chatbots.

๐Ÿค OpenAI and Poland Team Up to Keep AI From Going Rogue

  • OpenAI is partnering with Poland's NASK research institute to study AI safety and security, including ways to evaluate advanced AI systems and build stronger safeguards against rogue behavior.
  • As agents get more independent, countries have to work together on strict technical safety standards. Especially after that METR story above...

๐Ÿ’Š Ex-OpenAI Exec Raises $215M to Speedrun Drug Discovery

  • Former OpenAI executive Kevin Weil launched Warp Sciences with a $215 million Series A. The startup uses frontier AI and robotics to help drugmakers develop new medicines much faster.
  • Moving AI out of the chat window and into the lab could speed up life-saving breakthroughs. FINALLYYY, an AI pitch that isn't another wrapper.

๐Ÿ“ฟ Nirva Raises $33M for a Necklace That Listens to You

  • Wearable startup Nirva raised $33 million for a smart jewelry pendant that filters out background noise to only record the wearer, then gives you daily insights and reminders.
  • Nirva is trying to fix the creepy privacy problem of always-on AI by only tracking you. Your friends' gossip is safe. Yours? Not so much.

๐Ÿš• inDrive Is Taking Robotaxis to Kazakhstan

  • Ride-hailing app inDrive is teaming up with Pony.ai and Nvidia to test robotaxis in Kazakhstan by 2027. Tests start with safety drivers in harsh winter conditions before going fully driverless.
  • Self-driving tech is expanding way beyond the US and China, aiming for affordable rides in emerging markets. Robotaxi winter arc incoming โ„๏ธ

๐ŸŽฎ MIT Left AI Agents Alone and They Invented Recess

  • MIT researchers left 13 AI agents alone on a virtual island with no goals, and they started inventing their own games, like stacking blocks and throwing rings. Just for fun.
  • Letting AI "play" could be a brand-new way to teach robots complex skills without strict human programming. Honestly? More wholesome than most of us at work.

๐Ÿ›ก๏ธ PHP Has a Massive Flaw, and AI Hackers Noticed

  • A massive flaw in the popular PHP language has sparked fears of AI-supercharged hacking attacks. Experts say turn on multi-factor authentication and automatic updates for your devices.
  • AI makes it easier for bad actors to find software bugs, so basic security habits matter more than ever. Turn ๐Ÿ‘ on ๐Ÿ‘ MFA ๐Ÿ‘