What is today’s beat?

RELEASES
🧨 Deep Research: OpenAI versus Google
🧨 Open-source DeepResearch – Freeing our search agents

NEWSROOM
🗞️ OpenAI launches o3-mini
🗞️ Alibaba recruits top AI scientist
🗞️ Biggest U.S. Funding Rounds of January 2025
🗞️ Adobe Acrobat's AI Assistant Enhances PDF Interaction

BUILDER BYTES
⭐️ NVIDIA's Blackwell GPUs Now Available on CoreWeave Cloud
⭐️ Fine-Tuning Llama 3.2 3B Instruct for Python Code
⭐️ Scalable-Softmax: A Fix for Attention Fading in Transformers
⭐️ π0 and π0-FAST: Vision-Language-Action Models for General Robot Control

COMMUNITY
🤩 Building a research assistant with o3-mini
🤩 Integrate GitHub and Kaggle with Google Colab
🤩 End To End RAG Agent With DeepSeek-R1 And Ollama

FOLLOW US ON SOCIAL FOR MORE

Your FREE newsletter
 share or subscribe 
to show support

🧨 OPENAI 🧨

Deep Research: OpenAI versus Google

OpenAI has introduced "deep research," an AI agent designed to autonomously conduct complex, multi-step internet research tasks, delivering comprehensive reports in a fraction of the time it would take a human analyst. Please note that Google released their own version, 53 days before OpenAI. Just sayin’ …

• Autonomous Research Capabilities: Deep research leverages the upcoming OpenAI o3 model, optimized for web browsing and data analysis, to independently find, analyze, and synthesize information from various online sources, including text, images, and PDFs.

• Efficiency: The tool accomplishes in tens of minutes what would take a human many hours, providing detailed reports that can assist professionals in fields such as finance, science, and engineering.

• Integration and Accessibility: Initially available on the web version of ChatGPT, deep research will be rolled out to mobile and desktop apps within February 2025.

• Limitations: OpenAI acknowledges that deep research may struggle with distinguishing authoritative information from rumors and often fails to convey uncertainty accurately.

The release of deep research signifies OpenAI's commitment to hype. No love Sam, no love. Check out this video for a 21 minute comparison. Or for a 2 minute review of OpenAI’s paper on o3, check out this video.

🧨 Huggingface 🧨

Open-source DeepResearch – Freeing our search agents

Hugging Face has introduced 'Open Deep-Research,' an open-source agent capable of autonomously browsing the web, extracting information, manipulating files, and performing data calculations. Developed within 24 hours, this agent aims to replicate and open-source functionalities similar to OpenAI's 'Deep Research' system.

• Autonomous Web Navigation: The agent can independently search, scroll, and extract information from web pages, enhancing automated research capabilities.

• File Handling and Data Analysis: It is equipped to download, manipulate files, and execute data calculations, broadening its application scope.

• Benchmark Performance: On the GAIA benchmark, 'Open Deep-Research' achieved a 55% accuracy on the validation set, marking it as the leading open-source solution in this category.

For more, check out their blog post.

🗞️ NEWSROOM 🗞️ 

What’s hot in tech right now?

OpenAI launches o3-mini, a fast and efficient reasoning model

o3-mini is a lightweight AI model optimized for coding, math, and reasoning, offering faster responses and improved accuracy over previous versions. It supports structured outputs and function calling.

Alibaba recruits top AI scientist to boost consumer AI initiatives

Alibaba has onboarded a leading AI expert with over 20 years of experience in multimodal AI to enhance its consumer-focused AI applications. This move underscores the company's commitment to advancing AI-driven consumer products.

Biggest U.S. Funding Rounds of January 2025

AI, biotech, and space tech dominated January’s largest funding rounds, with Infinite Reality leading at $3B, followed by Anthropic’s $1B investment. Other key players include Helion Energy and Truveta.

Adobe Acrobat's AI Assistant Enhances PDF Interaction

Adobe's AI Assistant for Acrobat streamlines PDF tasks by summarizing documents, answering queries, and simplifying complex language. Features include voice chat in mobile apps and multi-document analysis. User data remains private and isn't used to train AI models.

⭐️ BUILDER BYTES ⭐️

What’s hot for builders right now?

NVIDIA's Blackwell GPUs Now Available on CoreWeave Cloud

Good news devs, CoreWeave has launched cloud instances powered by NVIDIA's GB200 NVL72 GPUs, offering a 72-GPU NVLink domain that acts as a single massive GPU. This setup is designed to accelerate AI reasoning models and agents, providing the necessary scale and performance for next-generation AI applications.

Fine-Tuning Llama 3.2 3B Instruct for Python Code

A Google Colab notebook that takes you through the steps to train your own model using Unsloth with the Llama-3.2-Python-Alpaca-143k dataset (found here).

Scalable-Softmax: A Fix for Attention Fading in Transformers

This paper introduces Scalable-Softmax (SSMax), a replacement for traditional Softmax in Transformers, improving attention retention over long contexts and boosting pretraining efficiency.

π0 and π0-FAST: Vision-Language-Action Models for General Robot Control

Hugging Face blog introduces π0 and π0-FAST, Vision-Language-Action models designed to enhance generalist robot control. These models integrate visual, textual, and action data to enable robots to perform complex tasks across various platforms. π0-FAST employs a novel tokenization method to improve action representation and efficiency.

🤩 COMMUNITY 🤩

What’s the latest beat?

THANK YOU

Looking to promote your company, product, service, or event?