Good morning. It’s Friday, January 31st.

Did you know: 54 years ago today, Ham the Chimp traveled into outer space aboard Mercury-Redstone 2?

In today’s email:

• OpenAI Poised for o3-mini Release

• Meta’s 2025 AI Roadmap

• Tesla’s 2025 AI Plans

• 2 New AI Tools

• Latest AI Research Papers

You read. We listen. Let us know what you think by replying to this email.

OpenAI Poised for o3-mini Drop as DeepSeek R1 Gathers Steam

OpenAI is ready to drop its latest reasoning models, "o3-mini" and "o3-mini-high," aimed squarely at tasks that require a bit more cognitive horsepower, like coding and scientific analysis.

The new models promise faster responses and better reasoning, outpacing o1 while improving cost efficiency. As DeepSeek R1 continues to make waves with surprising performance at a fraction of the cost, OpenAI is positioning o3-mini to counter this growing threat.

Expect the models to hit both API and ChatGPT, with Plus-tier users getting 100 queries a day. Read more.

Meta’s 2025 AI Roadmap Sets the Stage for Open-Source Dominance and Next-Gen Wearables

Meta CEO Mark Zuckerberg has laid out a high-stakes agenda for 2025, with Llama 4 taking centre stage. This omni-model, engineered for multimodal interactions and agentic functions, positions Meta as a serious contender in the open-source AI race, aiming to disrupt closed ecosystems. Meanwhile, the Meta AI assistant is on track to exceed one billion users, offering a tailored experience that adapts to user contexts, cultures, and individual preferences.

Meta’s AI-driven Ray-Ban smart glasses also hit a milestone, surpassing one million units sold in 2024. The next iteration, codenamed Hypernova, is set to feature an integrated heads-up display and next-gen AI capabilities, reinforcing the vision of AI glasses as a new computing platform.

On the infrastructure front, Meta is pouring resources into AI, deploying nearly a gigawatt of computational power and constructing a two-gigawatt data centre. By 2025, Zuckerberg expects the creation of an AI engineering agent with the skills of a mid-level engineer, which could shift the AI research landscape. Read more.

Elon Musk Details Bold AI Plans for Tesla and Optimus – ‘Things are really going to go ballistic next year’

Elon Musk painted a bold picture during Tesla’s Q4 2024 earnings call, forecasting exponential growth driven by AI and robotics. With a laser focus on self-driving tech and humanoid robots, Musk claims Tesla could soon surpass the combined value of the next five largest companies. Despite automotive margins shrinking, Tesla’s energy storage business soared, with growth projections exceeding 50% in 2025.

Musk aims to produce thousands of Optimus robots in the coming year, envisioning $10 trillion in revenue from the venture. Meanwhile, battery production is set to ramp up in 2025, addressing supply constraints, while the Tesla Semi is primed for a revenue boost by early 2026. Full transcript here. Read more.

• Apple secretly teams with SpaceX to add Starlink in iOS 18.3

• Dario Amodei Challenges DeepSeek’s $6M AI Hype

• SoftBank plans billion-dollar investment in OpenAI, Microsoft hosts Deepseek

• OpenAI brings its AI models to nuclear research at Los Alamos National Laboratory

• OpenAI to unveil new tech to D.C. power players

• Mistral, Ai2 release new open-source LLMs

• DeepSeek-R1 Now Live With NVIDIA NIM

• Google Quietly Rolls Out Gemini 2.0 Pro Experimental

• Intel has already received $2.2B in federal grants for chip production

• U.S. Copyright Office says AI generated content can be copyrighted — if a human contributes to or edits it

• Pika Labs Launches Turbo Mode: 3x Faster AI Video Generation at Lower Cost

• AI chipmaker Cerebras says it's been 'crushed with demand' for China's DeepSeek from business customers

• Deutsche Bank has published a bunch of memes about DeepSeek

• Ex-Google, Apple engineers launch unconditionally open source Oumi AI platform that could help to build the next DeepSeek

• OpenAI releases open-source voice-controlled solar system demo using WebRTC integration

• China’s secret laser nuclear fusion facility exposed by satellite

• New training approach could help AI agents perform better in uncertain conditions

• “Everything that moves will be robotic”: Nvidia CEO Jensen Huang says robots and self-driving cars are just around the corner

• Alibaba says its new Qwen2.5-VL model is a useful "visual agent"

• Luma AI Launches 4K Video Upscaling Feature for DreamMachine Platform

• KREA AI Launches Character Consistency Feature for Video Generation

• Gemini AI can automatically turn your spreadsheets into charts

• Figure AI humanoid robots to pass brutal battery, safety tests before launch

• ChatGPT's mobile users are 85% male, report says

📄 SFT Memorizes, RL Generalizes: A Comparative Study of Foundation Model Post-training

📄 Critique Fine-Tuning: Learning to Critique is More Effective than Learning to Imitate  ​​

📄 Exploring the sustainable scaling of AI dilemma: A projective study of corporations' AI environmental impacts

📄 GuardReasoner: Towards Reasoning-based LLM Safeguards

📄 Thoughts Are All Over the Place: On the Underthinking of o1-Like LLMs

Thank you for reading today’s edition.

Your feedback is valuable. Respond to this email and tell us how you think we could add more value to this newsletter.

Interested in reaching smart readers like you? To become an AI Breakfast sponsor, reply to this email or DM us on X!