- NATURAL 20
- Posts
- Demis Hassabis: AI's Biological Vision
Demis Hassabis: AI's Biological Vision
PLUS: Gemini to Lead Google’s AI Push, Meta Integrates AI in Social Media and more.
Conquer Winter in Style: Best Men’s Sneakers from Nike Air Max.
Winter is here, and it’s time to elevate your look while staying warm and comfortable. The Nike Air Max collection is perfect for the modern man ready to tackle the season. From the edgy Air Max Plus to the versatile AM1, these sneakers combine style with features designed to keep your feet happy all winter long.
Whether you're tackling holiday shopping or celebrating with friends, these kicks are your perfect companion. Find your ideal pair and effortlessly elevate your winter wardrobe this season.
Treat yourself or a loved one to a stylish pair from the collection this holiday season.
Today:
Demis Hassabis: AI's Biological Vision
DeepSeek-V3 Surpasses Llama & GPT
OpenAI Restructures 2025
Gemini to Lead Google’s AI Push
Meta Integrates AI in Social Media
BiMediX2 Excels in Medical Imaging
New Tests for Smarter AI Models
Google's "Era of Digital Biology" BEGINS…
Demias Habis, Nobel Prize winner in chemistry for his work on AlphaFold, envisions a "digital biology" era powered by AI. He highlights AI's ability to solve complex problems, such as protein folding, and its transformative potential in drug discovery, biology simulation, and scientific exploration.
With innovations like AlphaFold and the promise of AGI, Hassabis underscores the importance of responsibly harnessing AI to accelerate breakthroughs in biology and beyond.
Chinese AI startup DeepSeek launched DeepSeek-V3, an ultra-large open-source AI model with 671B parameters, designed for efficiency using a mixture-of-experts architecture. It outperforms Meta’s Llama-3.1 and rivals closed models like GPT-4o in most benchmarks. Key innovations include load balancing and multi-token prediction, boosting speed and accuracy. Training costs were significantly lower than competitors. Available via Hugging Face and DeepSeek Chat, the model pushes open-source AI closer to parity with proprietary systems.
OpenAI plans to transition into a for-profit Public Benefit Corporation (PBC) in 2025, putting its for-profit arm in charge while maintaining a nonprofit for charitable initiatives. This move aims to secure capital for advancing AI development, including artificial general intelligence. The change has sparked controversy, with critics like Elon Musk and Mark Zuckerberg opposing it. OpenAI asserts that the restructuring will balance profit with societal benefit and fuel AI innovation.
Google CEO Sundar Pichai emphasized 2025 as a pivotal year for AI development, naming the Gemini model and its consumer-focused app as the company’s top priority. Speaking at a strategy meeting, Pichai highlighted the need for faster innovation to secure leadership in the competitive AI market. While Gemini shows strong momentum, Pichai acknowledged challenges ahead, stressing the urgency to scale and improve as Google deepens its investment in AI technologies.
Meta plans to integrate AI characters into Instagram and Facebook, enabling them to generate and share content. These characters, equipped with bios and profiles, aim to enhance user interaction and creativity. Despite concerns about potential misuse, Meta enforces labeling rules for transparency. The company is also advancing AI tools for creators, including photo editing, AI assistants, and text-to-video generation, positioning AI as a core element of its social media strategy in 2025.
Researchers at Mohamed Bin Zayed University have developed BiMediX2, the first AI system capable of analyzing medical images in both English and Arabic. Trained on 1.6 million medical texts and images, it excels in processing X-rays, MRIs, and microscopic images, outperforming existing models, particularly in Arabic content. Built on Llama 3.1, BiMediX2 is available for research use, with tools like BiMed-MBench to support further bilingual AI advancements.
AI evaluations are evolving to keep pace with rapid advancements in artificial intelligence. New benchmarks, like Epoch AI's FrontierMath and METR's RE-Bench, assess capabilities such as mathematical reasoning and real-world problem-solving. These tests aim to uncover AI’s strengths and risks, including potential misuse. With AI labs releasing powerful models, experts emphasize the urgent need for rigorous, third-party evaluations to ensure safety, reliability, and preparedness for emerging challenges in this transformative field.
🧠RESEARCH
DiTCtrl introduces a training-free method for generating longer videos with multiple prompts using the Multi-Modal Diffusion Transformer (MM-DiT). It ensures smooth transitions and consistent motion across prompts by leveraging attention mechanisms. Paired with the new MPVBench benchmark, DiTCtrl delivers state-of-the-art multi-prompt video generation without additional training.
ReMoE introduces a fully differentiable Mixture-of-Experts (MoE) model using ReLU routing to replace traditional TopK routers. It improves performance, scalability, and load balancing across experts. Experiments show ReMoE outperforms traditional MoE, scaling efficiently with more experts while enabling domain specialization. Implementation is available on GitHub.
PartGen introduces a novel method for creating 3D objects with meaningful, editable parts from text, images, or unstructured 3D models. Using multi-view diffusion models, it segments and reconstructs cohesive 3D parts, outperforming existing methods. PartGen enables detailed 3D part editing and enhances creative workflows.
🛠️TOP TOOLS
Clip Interrogator - AI-powered tool that analyzes images and generates descriptive text prompts.
Pika AI - Transform text and images into captivating video content.
Stable WarpFusion v0.15 - AI-powered image manipulation tool that offers users the ability to create stunning visual effects and transformations.
ValidatorAI - AI-powered platform designed to assist entrepreneurs in validating and developing their startup ideas.
Screenshot to Code Converter - Converts screenshots into clean, functional code for HTML, Tailwind, React, or Vue.
📲SOCIAL MEDIA
I find that an exciting thought. Today's generation will no longer be able to live without generative AI. Just as the generation I grew up in could hardly answer a question without Wikipedia and Google, today's generation will be able to answer everything with generative AI.
— Chubby♨️ (@kimmonismus)
12:58 AM • Dec 27, 2024
What'd you think of today's edition? |
Reply