• The Tensor
  • Posts
  • OpenAI’s pivot to For-Profit for $150 Billion

OpenAI’s pivot to For-Profit for $150 Billion

Zuckerberg 2.0 - Greek revival to AI domination

Author

Dev Sharma
September 16, 2024

Welcome to today’s edition of The Tensor, get smarter with executive insights on the latest in AI & Tech Industry, 5 min reads, 3x a week.

In today's Tensor:

  • Top Story - OpenAI’s pivot to For-Profit for $150 Billion

  • Industry - Zuckerberg 2.0 - Meta’s path to dominate in AI

  • Research - Behind OpenAI’s o1 - How AI models learn to think before speaking

  • Quick Bytes - Amazon joins C2PA, Godmother of AI raises $230M for her startup, and 2 more.

Let’s dive in

Source: Bloomberg via Getty Images

The scoop: OpenAI has been loosely governed by its non-profit board, which essentially lost its teeth during last year’s CEO drama. Now, the company plans to remove the board entirely and shift to a for-profit model. Why?

To raise more money at a jaw-dropping $150B valuation. By removing profit caps, OpenAI hopes to attract heavyweight investors, with existing backers like Khosla Ventures and Microsoft expected to re-up, while new players like Nvidia, a UAE fund, and Apple might jump aboard. OpenAI’s decided to trade their solar powered private jet for a space shuttle.

Details:

  • Structural Overhaul: OpenAI will transition from a non-profit controlling a for-profit arm to a standard corporate structure, simplifying governance.

  • Funding Boost: They're raising cash through convertible notes, which is basically a financial IOU that turns into stock later, giving it a huge cash injection.

  • Cost Management: With the latest Nvidia super chip, the GB200, priced at around $80K each, training AI’s next top model is getting pricey. Their current billions in revenue aren’t enough to fund the next generation of AI.

Why it matters: OpenAI kickstarted the AI boom and has been leading the charge. Now, they’re signaling that to maintain that lead, they’re willing to shift focus from their original mission to benefit humanity and concentrate on achieving AGI. In an industry already overflowing with sky-high valuations, OpenAI’s $150 billion target should have seasoned investors doing a double-take. Given their existing revenue and profit-sharing deals with Microsoft, justifying this valuation feels like trying to sell sand in the desert—they’ll need something truly groundbreaking.

Bottom line: The AI race is far from over, with titans battling for supremacy. One thing’s clear: Nvidia seems to be the real winner here, as everyone’s lining up to buy their hardware—they’re the only shovel seller in a town full of miners. It'll be interesting to see if OpenAI’s gamble pays off.

The scoop: Mark Zuckerberg is rebranding—not just Meta, but himself. At 40, the billionaire has unveiled a more relatable persona. He appeared at the Chase Center for an interview with the "Acquired" podcast. Donning a casual outfit and making Greek references, he's aiming to change how billions interact with AI.

Highlights:

  • Meta’s Progress with Llama 3.1: With the release of Llama3.1, meta caught up with the leading closed frontier models like GPT4, and Zuckerberg has no plans to stop being the champion of open source, hinting at continued progress and same open approach.

  • Return to "Move Fast and Break Things": Echoing Facebook's early mantra, Zuckerberg highlighted a renewed focus on quick shipping and iterative development. He suggested that Meta is embracing speed over perfection, releasing products early to gather user feedback—even if they're not polished.

  • Lessons from Aeschylus: Sporting a shirt with ancient Greek text translating to "Learning through suffering," he drew inspiration from the playwright Aeschylus. This reference underscored his long-term vision and the growth that comes from overcoming challenges.

  • Brand Recovery Timeline: Trust will be a currency with which we interact with AI in the future, and acknowledging past missteps, Zuckerberg estimated it might take another 12 years for Meta's brand to fully recover from its 2016 downturn.

Bottom line: As one of the most powerful figures in tech, Zuckerberg's transformation impacts us all. Meta has an leading AI lab, billions in the bank and world's largest distribution platform. His new direction impacts how and what AI billions of users interact with. We're witnessing a veteran musician drop a surprise album that reinvents the genre—others adapt or risk becoming obsolete.

The scoop: OpenAI’s release of o1 which you can read in our previous edition has been built on recent breakthroughs that enable AI language models to self-reflect, reason, and explain their thought processes before generating outputs. Over the weekend, I dug into these papers and have complied research from OpenAI, Stanford, and Microsoft that is aiming to enhance AI transparency and reliability, making machines that not only provide answers but understand how they got there.

Different research methods:

  • Stanford's Quiet-STaR models allow AI to evaluate and correct their reasoning before responding.

  • Agent Q empowers AI agents to justify their actions and adapt to new situations effectively.

  • OpenAI's "Let's Verify Step by Step" breaks down complex tasks into smaller steps, each verified for correctness.

  • Microsoft/Mila's V-STaR trains AI to verify its own reasoning, ensuring accurate and reliable outputs.

  • Notre Dame/Tencent's reflective models integrate self-assessment in mathematical reasoning tasks.

Why it matters: Hallucination in AI models has been the biggest issue in wide enterprise adoption. These advancements in self-reflective AI models are crucial for developing trustworthy and transparent systems in high-stakes applications like healthcare, finance, and education, where errors could have severe consequences.

Bottom line: Just as a student double-checks their homework to catch mistakes, these AI models review their own thinking to ensure accuracy. They're not just getting smarter—they're learning how to think critically, making them more reliable partners in tackling complex problems.

Thank you for reading today’s edition of the Tensor!
Someone forwarded you today’s edition and you want to read more?