Welcome back to Daily Zaps, your regularly-scheduled dose of AI news ⚡️ 

Here’s what we got for ya today:

• 🗺️ OpenAI releases "economic blueprint" for AI

• ♼ Zuck says AI replacing "mid-level engineers" by 2025

• 👾 White House announces 'final rule' on AI chip exports

• 🤖 Open source AI reasoning model for $450

Let’s get right into it!

Sign Up | Advertise | Tools

STARTUPS

OpenAI releases "economic blueprint" for AI

OpenAI has released an "economic blueprint" for AI, outlining policies to collaborate with the U.S. government and allies to maintain global AI leadership. The document, featuring input from Chris Lehane, OpenAI’s VP of global affairs, calls for increased federal investment in chips, energy, and talent while streamlining regulations to address conflicting state-level AI laws. It criticizes the CHIPS Act’s inefficiency and highlights the need for expedited infrastructure development, including energy sources like nuclear power, to support AI growth.

The blueprint advocates for voluntary government-industry collaboration on AI model safety, export controls to allies, and international standards. It also addresses contentious copyright issues, urging the U.S. to enable AI training on publicly available information to remain competitive globally. OpenAI’s lobbying efforts and partnerships with U.S. government agencies, such as the Pentagon, underline its intent to shape U.S. AI policy while navigating potential legislative hurdles, including challenges to the Biden administration's AI Executive Order.

The full Blueprint is available here.

BIG TECH

AI replacing "mid-level engineers" by 2025

Mark Zuckerberg announced that Meta plans to automate the work of midlevel software engineers by 2025, with AI eventually taking over all coding for its apps. While initially costly, Zuckerberg envisions AI-generated code streamlining operations and reducing reliance on high-paid engineers.

“I think this year probably in 2025, we at Meta as well as the other companies that are basically working on this are going to have an AI that can effectively be a sort of midlevel engineer that you have at your company, that can write code and once you have that, then in the beginning it'll be really expensive to run and you can get it to be more efficient and then over time we'll get to the point where a lot of the code in our apps and including the AI that we generate is actually going to be built by AI engineers instead of people engineers.''

Full interview here.

GOVERNMENT

White House announces 'final rule' on AI chip exports

The Biden administration has proposed a new framework for exporting advanced AI chips, aiming to balance national security with economic interests. The plan restricts chip exports to over 120 countries, including Mexico and Israel, while maintaining access for about 20 key allies. Focused on limiting China’s AI capabilities, the rules seek to close loopholes in existing controls, but industry leaders warn they could harm global competitiveness and innovation, particularly by restricting widely used gaming and data center chips.

With a 120-day comment period, the policy’s future will depend on the incoming Trump administration, which faces the challenge of navigating U.S.-China trade tensions while defining its AI strategy.

BIG TECH

Open source AI reasoning model for $450

Reasoning AI models are becoming more accessible and affordable, as demonstrated by the release of Sky-T1-32B-Preview by NovaSky, a UC Berkeley research team. Sky-T1, trained for just $450, is a competitive open-source reasoning model, with its dataset and training code available for replication. It outperforms an earlier version of OpenAI’s o1 on key benchmarks like MATH500 and LiveCodeBench, although it lags behind on physics and biology-related GPQA-Diamond questions.

Leveraging synthetic data and tools like Alibaba's QwQ-32B-Preview and OpenAI’s GPT-4o-mini, the model was trained efficiently using 8 Nvidia H100 GPUs. While OpenAI plans to release more advanced models soon, NovaSky aims to refine their open-source approach, focusing on creating more efficient, high-performing reasoning models.