Training a top AI reasoning model now costs less than $450

Hi everyone! Here’s what you need to know about AI today:

👉 Training a top AI reasoning model now costs less than $450

👉 OpenAI’s new goal is to beat China in AI

👉 Nvidia CEO praises Elon Musk for his AI efforts

and many more!

📧 Did someone forward you this email? Subscribe here for free to get the latest AI news everyday!

Read time: 4.9 minutes

What’s going on: NovaSky which is a team of researchers from UC Berkeley’s Sky Computing Lab, released a reasoning model, Sky-T1, that has shown strong competitiveness with one of the earlier versions of OpenAI’s o1 on some key benchmarks. The model is the first open-source ‘reasoning’ model that can be trained from scratch and it only costs less than $450 to train.

What does it mean: It usually takes hundreds of thousands of dollars to train AI models and in some cases even millions! The fact that it is possible to build AI models with high-level reasoning capabilities at this level of affordability and efficiency is turning a lot of heads. Synthetic data, or training data generated by other models, has helped the NovaSky team to achieve this milestone.

More details:

• The NovaSky team used another reasoning model, Alibaba’s QwQ-32B-Preview, to generate the initial training data for Sky-T1, then curated the data mixture and utilized OpenAI’s GPT-4o-mini to reformat the data.

• Training the 32-billion-parameter Sky-T1 took about 19 hours using 8 Nvidia H100 GPUs.

• Sky-T1 performs better than OpenAI’s o1 on solving competition-level math challenges and it also beats earlier versions of o1 on a set of difficult problems from LiveCodeBench benchmark.

• Palmyra X 004, a model recently released by AI company Writer, was trained almost entirely on fake data and cost just $700,000 to develop.

What’s going on: OpenAI has shared a bunch of ideas on how AI should be managed and competed with. They've put together what they call an "economic blueprint" to shape how new technology policies will be made. This comes at a time when a new administration is starting, and Elon Musk, who has had public disagreements with OpenAI, is influencing policy in Washington.

What does it mean: OpenAI wants to make sure they have a big voice in how AI is controlled in the US, especially with new leaders in charge. They're aiming to keep AI development under American influence, not letting countries like China take the lead. They also want to make sure that the benefits of AI reach all parts of the US, not just big tech cities, creating jobs and opportunities everywhere.

More details: 

• OpenAI propose that AI companies should share their compute resources with public universities.

• They believe US companies should ease the export of advanced "frontier AI" models to America’s allies and partners so they can "stand up their own AI ecosystems" based on US rather than Chinese technology.

• OpenAI also thinks that we might ensure AI "has the ability to learn from universal, publicly available information, just like humans do, while also protecting creators from unauthorized digital replicas."

🤖 OpenAI is reviving its robotics plans with new job listings that focus on developing "general-purpose" adaptive robots equipped with custom sensor suites for real-world settings.

📄 VCs believe that for AI startups to stand out in a crowded market, they must secure access to proprietary and high-quality data sets and not just work hard to make more sophisticated AI models. The data could come from exclusive partnerships, industry-specific insights, or novel data collection methods.

👏 Nvidia CEO Jensen Huang praised Elon Musk for his work on AI, highlighting xAI's cognitive intelligence, Tesla's autonomous vehicles, and the Optimus humanoid robot as key areas of focus.

⚠ Geoffrey Hinton, the 'Godfather of AI' and a 2024 Nobel Prize winner in Physics, warns about the potential risks of AI, expressing concerns that advanced AI could surpass human intelligence and control.

👨‍💻 FunkSec, an AI-powered ransomware group, has attacked 85 organizations worldwide, using AI for both hacking and demanding small ransoms, while mixing activism with crime.

🏰 The UK government has launched an AI Opportunities Action Plan, aiming to make the country a global AI leader with initiatives like AI Growth Zones and a significant increase in computing power.

🏭 Coreweave, the $19 billion cloud computing giant that provides companies with AI compute resources, has formally opened its first two data centers in the UK.

Improve your grammar

This prompt effectively improves your writing by guaranteeing correct English grammar and eliminating spelling, punctuation, and grammatical mistakes.

Vitol - Senior Machine Learning Engineer

Writer - AI Engineer

Statsig - Data Scientist- Growth

Point72 - Data Scientist, Proprietary Research

Thank you for staying with us like always! If you are not subscribed, subscribe here for free to get more of these emails in your inbox! Cheers!