DeepSeek Outperforms DALL-E 3 and Stable Diffusion

Hey there 👋

We hope you're excited to discover what's new and trending in AI, ML, and data science.

Here is your 5-minute pulse...

print("News & Trends")

DeepSeek releases Janus-Pro: an open-source model that outperforms DALL-E 3 and Stable Diffusion

DeepSeek unveils Janus-Pro, an open-source multimodal AI with superior text-to-image generation and understanding. It outperforms competitors, scales up to 7B parameters, and refines visual encoding for better accuracy. Free under MIT license, now available on Hugging Face.

Haptikos unveils a hand exoskeleton for grabbing digital things in XR (7 min. read)

Haptikos has developed a hand exoskeleton that allows for precise haptic interactions in extended reality (XR) devices. It raised $2.7 million in funding and plans to eventually offer a consumer version. The exoskeleton works with Haptikos' AI-powered operating system and has applications in various industries. It has been tested

OpenAI has evidence that its models helped train China’s DeepSeek (2 min. read)

Chinese AI company DeepSeek disrupted Silicon Valley with cheaply developed AI models, possibly built using data from OpenAI. Microsoft and OpenAI are investigating. OpenAI suspects DeepSeek used distillation, a violation of terms. OpenAI defends its IP and works with US government to protect against competitors.

The U.S. Copyright Office releases key guidelines for understanding AI Copyright

The U.S. Copyright Office confirms AI-generated content alone isn’t copyrightable—only human-authored elements qualify. Extensive prompting doesn’t count. Policies remain unchanged, reinforcing that meaningful human creativity is essential for copyright protection in AI-assisted works.

print("Applications & Insights")

Extracting Structured Vehicle Data from Images (8 min. read)
Build an automated vehicle documentation system using GPT-4, LangChain, and Pydantic. Extract structured data—type, plate, make, model, and color—from images efficiently. This flexible pipeline transforms raw images into actionable insights. Full tutorial and Colab notebook included

Using DeepSeek-R1 Locally (5 min. read)
DeepSeek's R1 model rivals OpenAI's o1 in performance and is free to use. Set it up locally with Docker and Ollama for offline access. Even on an 8GB RAM laptop without a GPU, it runs efficiently.

How to deploy and fine-tune DeepSeek models on AWS (7 min. read)
DeepSeek AI released their DeepSeek-R1 model, breaking the internet and stock market. Collaborating with AWS, they make it easier to deploy and fine-tune models with Hugging Face on AWS services. Distilled models and LLMs are available in the Inference Endpoint Model Catalog, with optimized configurations for deployment. DeepSeek-R1-D

NLP Illustrated, Part 3: Word2Vec (13 min. read)
Part 3 of the NLP journey discusses the use of word embeddings, specifically the Word2Vec model, which uses intuitive math and machine learning to capture the nuances and relationships between words. The Skip-gram method is explained as the most commonly used approach.

print("Tools & Resources")

TRENDING AI TOOLS

• 🧠 Cerebrium: Deploy and scale AI models effortlessly.

• ⚡ MinusX: AI Data Scientist for Jupyter.

• 🐼 PandaETL: Lightweight ETL for seamless data processing.

• 📊 SheetAI: AI-enhanced data analysis in Google Sheets.

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 pulse experience for you.

See you soon,

Andres