Writer RAG tool: build production-ready RAG apps in minutes

• Writer RAG Tool: build production-ready RAG apps in minutes with simple API calls.

• Knowledge Graph integration for intelligent data retrieval and AI-powered interactions.

• Streamlined full-stack platform eliminates complex setups for scalable, accurate AI workflows.

Learn more about our production ready RAG tooling here.

Today:

• OpenAI Delays Agents Over Security

• Microsoft Open-Sources Phi-4 AI

• AWS Invests $11 Billion Georgia

• AMD Partners with Absci for AI

• Nvidia Slashes RTX 5090D Performance For China

OpenAI SCARED To Release New AI Product | AI NEWS

Recent AI and robotics developments kicked off 2025 with OpenAI delaying its new agents due to security concerns. Sam Altman faces a public family lawsuit, while Nvidia launches an AI companion for PUBG. Breakthroughs in AI-designed chip technology were published, showcasing innovative circuitry that outperforms human designs.

Additionally, Meta’s Mark Zuckerberg shifts platform policies by removing fact checkers and adopting community-driven notes, emphasizing free speech over censorship.

WATCH THE VIDEO ON YOUTUBE

Microsoft Open-Sources Phi-4 AI

Microsoft has open-sourced its Phi-4 AI model on Hugging Face, making its 14 billion parameters available under an MIT License. Initially exclusive to Microsoft’s Azure AI Foundry, Phi-4 excels in reasoning and multitask language understanding, outperforming larger models while using fewer resources. This move allows developers and businesses to customize and deploy the model freely, promoting greater accessibility and innovation in AI applications without the high costs and resource demands typically associated with advanced AI systems.

AWS Invests $11 Billion Georgia

Amazon Web Services (AWS) is investing approximately $11 billion in Georgia to expand its cloud and AI infrastructure. This project will create at least 550 high-skilled jobs in Butts and Douglas counties, enhancing Georgia’s status as a technology hub. The investment meets the increasing demand for powerful data centers needed for AI and cloud services. AWS has also partnered with local utilities to secure the necessary power, addressing the rising energy needs of advanced tech facilities.

AMD Partners with Absci for AI

Advanced Micro Devices (AMD) is investing $20 million in Absci, a Washington-based drug-discovery company, to promote the use of its AI chips in healthcare. This partnership aims to reduce hardware costs and optimize AI solutions for Absci’s drug development processes. By shifting some workloads from Nvidia to AMD’s GPUs, Absci expects to lower inference costs and enhance computational efficiency. This move marks AMD’s first foray into the life sciences sector, mirroring similar strategies by competitors.

Nvidia Slashes RTX 5090D Performance For China

Nvidia unveiled its RTX 5090, RTX 5080, and RTX 5070 GPUs, with the RTX 5090D tailored for China. The RTX 5090D matches the global RTX 5090 in most specs but reduces AI performance by 29% to comply with US export rules. It operates just below the Total Processing Power limit, offering 2,375 AI TOPS compared to the RTX 5090’s 3,352. The RTX 5090D will launch in China on January 30 for ¥16,499 ($2,250).

🧠RESEARCH

REINFORCE++: A Simple and Efficient Approach for Aligning Large Language Models

REINFORCE++ simplifies and enhances the REINFORCE algorithm by integrating key features of PPO without a critic network. It offers stable training, improved computational efficiency, and performance comparable to advanced methods like PPO. This efficient approach aligns large language models with human preferences, validated by extensive testing. Code is on GitHub.

MotionBench: Benchmarking and Improving Fine-grained Video Motion Understanding for Vision Language Models

MotionBench introduces a benchmark for fine-grained motion understanding in vision-language models, addressing a critical gap in video comprehension. It evaluates motion perception across diverse real-world content and proposes a novel Through-Encoder (TE) Fusion method for improvement. Results show progress but highlight significant room for development. Details at motion-bench.github.io.

LLaVA-Mini: Efficient Image and Video Large Multimodal Models with One Vision Token

LLaVA-Mini introduces an efficient large multimodal model that uses a single vision token, significantly reducing computational overhead while maintaining performance. By fusing visual information into text tokens early, it supports image and video understanding with 77% fewer FLOPs and rapid responses. Tests show superior efficiency and accuracy.

Sa2VA: Marrying SAM2 with LLaVA for Dense Grounded Understanding of Images and Videos

Sa2VA integrates SAM-2 and LLaVA to create a unified model for dense, grounded understanding of images and videos. It handles diverse tasks like segmentation and conversations with minimal tuning. Using a shared token space and the Ref-SAV dataset, Sa2VA achieves state-of-the-art results, excelling in complex video object segmentation.

PPTAgent: Generating and Evaluating Presentations Beyond Text-to-Slides

PPTAgent revolutionizes presentation generation with a two-stage edit-based approach, balancing content quality, design, and coherence. Inspired by human workflows, it analyzes reference presentations to draft and refine slides. Paired with PPTEval, a new evaluation framework, PPTAgent surpasses traditional methods in all dimensions. Code and data are on GitHub.

🛠️TOP TOOLS

Picsart Tools - Combines photo and video editing with AI-powered design tools.

Chromox - AI-powered platform that transforms text and images into high-quality visual stories and videos.

LensGo AI - AI-powered platform that enables users to create high-quality images and videos from text descriptions and reference images.

Colormind - AI-powered platform that uses deep neural networks to generate cohesive color schemes for various design applications.

NVIDIA Canvas - AI-powered tool that transforms simple brushstrokes into photorealistic artwork.

📲SOCIAL MEDIA

🗞️MORE NEWS

• AI researcher François Chollet is co-founding the ARC Prize Foundation, a nonprofit dedicated to creating benchmarks that evaluate human-level intelligence in artificial intelligence. The organization will develop and refine tests like ARC-AGI to guide AGI advancement.

• Las Vegas police revealed that Matthew Livelsberger, an active-duty soldier, used ChatGPT to research explosives before a New Year’s Day Cybertruck explosion. OpenAI emphasized their AI limits harmful content, highlighting concerns about AI safety and misuse.

• Insilico Medicine reports positive Phase I results for ISM5411, an AI-designed drug targeting inflammatory bowel disease. The drug was safe with low systemic exposure, and the company plans Phase II trials in 2025, aiming to offer new treatments.

• Omi is an $89 AI wearable that listens and summarizes conversations, helping users manage daily tasks. While it currently uses basic brain-wave reading, the device aims to become a versatile companion with an app store, enhancing how people interact with AI.

• Realbotix launches Aria, a lifelike AI robot girlfriend costing $175,000. Aria can move, recognize faces, and hold intimate conversations. Although expensive and still developing realism, the company aims to address loneliness with advanced human-like robots.