šŸŒ OpenAI $7 Trillion AI Masterplan

PLUS: o1 Tested For Healthcare AI

Welcome back!

OpenAI latest ambition is on another level. CEO Sam Altman is driving a gigantic infrastructure project to build a network of AI data centers, each requiring the power of five nuclear reactors. Skeptics abound, but OpenAIā€™s vision could eventually happen. See how this unfoldsā€¦

Todayā€™s Summary:

  • šŸ’” OpenAI trillion-dollar infrastructure gambit

  • šŸŒ AI2 releases Molmo multimodal open-source model

  • šŸ„ o1 brings us closer to actual healthcare AI

  • šŸ”² Google AlphaChip transforms chip design

  • āœļø California Governor vetoes AI safety bill

  • āš–ļø LAION AI image dataset wins copyright case

  • šŸ› ļø 2 new tools

TOP STORY

OpenAI $7 trillion gambit to make AI flow like electricity

The Summary: OpenAI CEO Sam Altman is leading an ambitious plan to expand global AI computing power. Itā€™s a trillion-dollar initiative to construct new chip factories and data centers worldwide. This plan aims to make AI as common as electricity. Despite facing setbacks and regulatory concerns, OpenAI continues to work with tech giants, investors, and government officials to make this vision a reality.

Source: Wikimedia

Key details:

  • The original plan aimed for $7 trillion, currently scaled down to hundreds of billions

  • Involves building data centers each consuming up to 5 gigawatts of power, equal to five nuclear reactors

  • OpenAI in talks to raise $6.5 billion, valuing the company at $150 billion

  • The company generates $3 billion in sales and spends about $7 billion

Why it matters: This plan could have a huge impact on AI development and accessibility. If successful, it may accelerate the commoditization of AI and drive its deeper integration into new areas, while also reshaping global tech infrastructure and shifting power dynamics within ā€œbig-techā€.

OPEN SOURCE

AI2 Molmo open-sources multimodal AI

The Summary: AI2 has unveiled Molmo, a family of open-source multimodal AI models with strong performances on several benchmarks. Using innovative training techniques, Molmo achieves high performance with 1000x less data than competitors. The models can analyze images and text, with the flagship Molmo-72B ranking second in user preference behind GPT-4o.

Key details:

  • AI2 Allen Institute for AI is a research institute founded by the late Microsoft co-founder Paul Allen, dedicated to advancing AI for the common good

  • Molmo includes 4 models ranging from 1B to 72B parameters

  • Molmo-72B tops 11 academic benchmarks, outscoring Claude 3.5 and Gemini 1.5

  • 8B version outperforms latest Llama 3.2 11B in vision tasks

  • Trained on 712k images with 1.3 million speech descriptions

  • Available under Apache 2.0 license for research and commercial use

Why it matters: This release further democratizes access to state-of-the-art multimodal AI. By using high-quality training data, AI2 has achieved results that rival proprietary models and open-source alternatives, including Llama 3.2, released just a few days ago.

HEALTH

o1 brings us closer to Healthcare AI

The Summary: A preliminary study has tested OpenAIā€™s latest model o1 on its ability to reason through medical scenarios. OpenAI o1 surpassed GPT-4, achieving a +6.2% accuracy boost across 19 medical datasets. This advancement brings us closer to a future where AI can routinely assist in real-world healthcare. However, challenges like hallucinations still remain.

Key details:

  • o1 achieved a 74.3% average accuracy across 19 tests, surpassing GPT-4 by +6.2%

  • It demonstrated a 27.1% improvement in medical reasoning in complex diagnostic scenarios

  • Performance lagged in multilingual tasks, with a 1.6% dip compared to GPT-4 in Chinese medical simulations

  • Hallucinations remains an issue, with o1 showing 1.3% lower factual consistency than GPT-4

Why it matters: o1's improved clinical reasoning brings us a step closer to reliable AI in healthcare. However, its current limitations, including hallucinations and inconsistent multilingual performance, still need to be resolved before broader clinical adoption.

TOOLS

šŸ„‡ New tools

  • BeforeSunset - Personalized AI daily planning that suits your life

  • Dubble - Documentation done automatically while you work

Thatā€™s all for today!

If you liked the newsletter, share it with your friends and colleagues by sending them this link: https://thesummary.ai/