Sign Up | Advertise | Podcast | AI University | | | Welcome, AI enthusiasts. | While tech giants continue to pour billions into AI development, a Chinese startup just proved you can build an open-source powerhouse on a tiny budget. | DeepSeek's V3 model competes with the top models at a fraction of the cost — and the walls between open and closed AI are crumbling faster than ever. Let's get into it… | | In today's AI rundown: | DeepSeek-V3 rewrites open-source AI playbook OpenAI reveals restructuring plans for next AI phase Turn your images into any style Stanford AI brings natural gestures to digital avatars 5 new AI tools & 5 new AI jobs More AI & tech news
| Read time: 4 minutes |
|
| | Friendly reminder: We're moving to a new email sending address. Make sure to add news@daily.therundown.ai to your contacts to ensure you keep receiving our emails. |
|
| | | | DEEPSEEK | |  | Image source: DeepSeek |
| The Rundown: Chinese AI startup DeepSeek just released DeepSeek-V3, a new powerhouse language model that sets new benchmarks in the open-source AI world with performance rivaling industry giants at a fraction of the cost. | The details: | V3 uses a Mixture-of-Experts architecture and maintains speed and cost-effectiveness despite its massive 671B parameter size. The training was completed in just two months at an estimated $5.57M, dramatically less than the reported $500M+ spent on models like LLaMA 3.1. The model shows exceptional strength in math and Chinese language tasks while matching or exceeding closed models across most benchmarks. V3 has been critiqued for identifying as ChatGPT in conversations, which may be due to significant GPT-generated content used in its training dataset.
| Why it matters: The gap between open and closed AI models has never been smaller. Chinese models continue to prove that the U.S. chip restrictions are failing to slow progress, and V3's benchmarks show that open-source, high-performance models are achievable without the massive resources of other tech giants. |
|
| | TOGETHER WITH VANTA | | | The Rundown: Join Vanta for a fireside chat with CEO Christina Cacioppo and "The Lean Startup" author Eric Ries to gain insights on navigating product-market fit, delegation, and scaling startups in the age of AI. | The Jan. 14 session will cover: | Key takeaways from Christina and Eric's entrepreneurial journeys How the principles of "founder mode" and lean methodology intersect The ways AI is transforming early-stage growth and startup success
| Secure your spot today. |
|
| | OPENAI | |  | Image source: Getty Images |
| The Rundown: OpenAI officially published its plans to transform into a public benefit corporation, posting a blog detailing the company's controversial shift from its nonprofit structure to enable massive funding abilities for future AI development. | The details: | The restructuring will convert OpenAI's for-profit arm into a Delaware-based PBC, with the original nonprofit gaining 'significant' shares in the new entity. OpenAI said the result would be 'one of the best-resourced non-profits in history,' enabling the pursuit of charitable goals in health care, education, and science. The process follows the startup's $6.6B funding round at a $157B valuation, which reportedly hinged on the restructuring. Elon Musk sued OpenAI in December to prevent the move, with California nonprofit Encode also pushing for a pause in a new filing.
| Why it matters: OpenAI's convoluted structure has been criticized since the Nov. 2023 board drama, but as Elon and co. continue to hold the startup's feet to the fire, the restructure process seems imminent. Threading the needle between a mission-driven focus and a for-profit endeavor is about to become a delicate balancing act. |
|
| | AI TRAINING | | | The Rundown: Google's new Whisk tool allows you to create custom-stylized versions of any photo by combining subjects, scenes, and artistic styles without writing prompts. | Step-by-step: | Visit Google Labs and open the Whisk experiment (free to use). Click "Start from scratch," upload your subject and scene, and choose a style. Let AI generate multiple consistent variations of your image. Refine results by adding specific details through the text box.
| Pro tip: Use the "+" button to add multiple subjects or scenes for more complex compositions. |
|
| | PRESENTED BY INNOVATING WITH AI | | | The Rundown: Innovating with AI's new program, AI Consultancy Project, transforms AI enthusiasts into professional consultants — tapping into a market projected to reach $54.7B by 2032. | The 6-month program delivers: | Proven frameworks for client acquisition and service delivery A step-by-step path to six-figure consulting income Students who land their first AI client in as little as 3 days
| Click here to request early access to The AI Consultancy Project. |
|
| | AI RESEARCH | |  | Image source: Stanford University |
| The Rundown: Stanford researchers just developed an AI model that enables digital avatars to produce natural, context-appropriate gestures while speaking — marking advances in making virtual characters more lifelike and expressive. | The details: | Training data included 1,000 hours of audiobooks and 60 hours of motion capture to understand relationships between speech, text, and body language. The system can generate appropriate gestures from text or audio input while detecting emotions from motion patterns alone. Using a new pre-training approach, the AI requires less training data than past models while achieving better performance on gesture generation benchmarks. The model shows particular strength in matching gestures to speech patterns, such as emphasizing specific words through corresponding body movements.
| Why it matters: Zoom CEO Eric Yuan has said we will soon send avatars to meetings in our place. Having realistic motion and gestures is a crucial step towards crafting truly reality-bending characters. Digital avatars are about to become ubiquitous across industries, and these advances could open up a whole new world of use cases. |
|
| | | | | ⚙️ ReactAI - Create fully-functional React components in seconds with no API key required 🔎 Hika - AI-powered knowledge search tool that transforms exploration and simplifies complex topics into interactive sections 📚 Scourhead - Free, open-source AI agent that scours the web, organizes data, and delivers results in a spreadsheet 🤖 SEObot - Produce useful, non-spammy AI-generated blog content with auto-linking, keyword research, image generation, and fact-checking 📊 Shoonya - Specialized foundation models fine-tuned for commerce use cases
| | 🤖 Meta - Software Engineering Manager - FAIR 🛡️ Shield AI - Product Marketing Manager Consultant 🚗 Waymo - Senior Program Manager, Commercial Operations 🔬 Scale AI - SEAL Research Scientist, Scalable Oversight 🏙️ Metropolis - Principal Software Engineer
|
|
| | | | 'Godfather of AI' Geoffrey Hinton raised his odds of human extinction risk due to superintelligent AI development, revising his estimate to 20% within three decades while calling for stronger government regulation of the technology. | OpenAI and Microsoft reportedly agreed on a concrete metric for achieving artificial general intelligence, with a newly revealed document from 2023 defining AGI as an AI system capable of generating $100B in annual profits. | Meta unveiled a vision for AI-generated characters to become active social media users across its platforms, with plans to expand capabilities from profile creation to content generation and live interactions. | Chinese robotics firm Unitree demoed B2-W, a rideable robot dog capable of carrying humans across challenging terrain in addition to acrobatic maneuvering and stability control. | Toyota's AI-powered humanoid robot CUE6 set a new Guinness World Record for the longest basketball shot by a robot, sinking an 80-footer on its second try. |
|
| | | | That's it for today!Before you go we'd love to know what you thought of today's newsletter to help us improve The Rundown experience for you. | | See you soon, | Rowan, Joey, Zach, and Alvaro—aka The Rundown Team | |
|
|
|
Tidak ada komentar:
Posting Komentar