Good morning. It’s Monday, April 3rd.

ChatGPT was banned in Italy over “data privacy” concerns. Reporter Peter Doocy was dismissed with laughter by the US Press Secretary for asking about a doomsday AI prediction at White House Briefing. Watch the video here.

In today’s email:

• Learning Vector Embeddings

• Mastering AI Image Generation

• BloombergGPT is AI for finance

• AI Tool of the Week: Wiseone

• Top Research Papers from the past 7 days

You read. We listen. Share your feedback by replying to this email, or DM us on Twitter. Also, we’ve launched a Podcast! Click below to listen to the latest episode, where Luke Wescott interviews AI-startup founder MJ on how he developed his AI music generator “Lemonaid.”

Learning AI: Vector Embeddings

If you could only know one thing about how AI works, let it be this:

In the rapidly evolving world of artificial intelligence, one concept has emerged as a cornerstone for understanding complex relationships and revolutionizing how programs like ChatGPT process information: vector embeddings.

Vector embeddings revolutionize language understanding by capturing semantic relationships in high-dimensional spaces, enabling AI to uncover meaning beyond mere keywords.

Vector embeddings allow us to represent words, phrases, and even entire documents as points in a high-dimensional space, capturing the semantic essence of language in a form that can be easily processed by LLMs. By measuring the distances between these points, LLMs can uncover the intricate connections and relationships between different linguistic elements, transcending traditional keyword matching and revealing the underlying meaning behind texts.

Example:

Imagine you're in geometry class, and you have a one-dimensional line. You can place a point on that line, say, three units from the origin, and another point, let's say, seven units from the origin. You can easily calculate the distance between these two points, which in this case is four units.

Now, if we move to two dimensions, we have two numbers that describe every point, and we can still calculate the distance between them using a ruler or math. The same concept applies to three-dimensional space, where three numbers describe every point. This is something we can easily understand because we live in a three-dimensional world.

Here’s where it gets abstract - and this is why ChatGPT works as well as it does:

Let's imagine a world with a thousand dimensions.

This means there are a thousand numbers that describe any particular point in this 1000-dimensional space. Now, what if we could represent every piece of text, like a blog post or a tweet, as a point in this 1000-dimensional space? We could reduce the meaning of these texts to a vector, a set of a thousand numbers that represent a point in this space.

Just like in one, two, or three-dimensional spaces, we can calculate the distance between these points. But instead of physical distance, we are measuring semantic distance, which tells us how related the meanings of two texts are, even if they use completely different words.

This concept is vector embeddings in a nutshell, and it allows us to find relationships between texts in a way that goes beyond simple keyword matching.

It's like taking Google Search and applying it to every other industry, enabling people to find connections between concepts like never before. The vector embeddings approach is revolutionizing how we analyze and process information, leading to a deeper understanding of the world around us.

_{sponsored post}

Master AI Image Generation

Develop a professional-level mastery of photorealistic AI-generated images with the new Dreambooth Mastery Course from PromptHero.

This advanced course teaches you how to train your own AI models to replicate any style, including photorealism and popular artists.

Master the use of Dreambooth + Stable Diffusion to create breathtaking images that are indistinguishable from real-life photographs.

Learn how to generate your own AI avatars and explore advanced techniques like Textual Inversion.

With the guidance of industry experts, you will learn to create personalized models that give you a competitive edge. Break free from the limitations of Stable Diffusion and supercharge your AI workflow.

This course is perfect for artists, creatives, and professionals from fields such as interior design, architecture, photography, and writing.

Discover how to harness the power of AI to enhance your work and stand out from the competition.

You'll get a comprehensive understanding of:

• model fine-tuning

• dataset creation

• data engineering

• cloud-based AI setups

Plus, you'll get an insider's look at how Openjourney was created.

No coding experience? No problem. The course is designed to be accessible even for those with limited technical knowledge.

Take your AI image generation skills to the professional level.

AI meets Finance

This week, Bloomberg unveiled BloombergGPT™, a large-scale generative AI model specifically designed to handle the complex and specialized language of finance.

This proprietary large language model (LLM) aims to harness the power of AI to transform the financial landscape," said Shawn Edwards, Bloomberg's Chief Technology Officer. "The model's focus on the financial domain will enable us to tackle new applications while delivering higher performance and faster time-to-market than custom models for each task."

The financial industry's unique challenges and terminology necessitate a domain-specific model, as generic LLMs struggle to achieve the same level of accuracy and effectiveness.

To do so, the team created a training corpus of over 700 billion tokens by combining a 363 billion token dataset from English financial documents with a 345 billion token public dataset. This data was used to train a 50-billion parameter decoder-only causal language model, which demonstrated impressive performance across finance-specific and general-purpose NLP tasks.

Gideon Mann, Head of Bloomberg's ML Product and Research team, emphasized the importance of quality data in the development of AI models: "Thanks to the collection of financial documents Bloomberg has curated over four decades, we were able to create a large and clean, domain-specific dataset to train a LLM that is best suited for financial use cases."

Read more: Full Research Paper

AI Tool of the Week: Wiseone

Wiseone is a new browser extension designed to optimize the online reading experience.

Browser extension features:

• Cross-checking: Allows users to verify facts by cross-referencing different sources on the same topic, promoting a well-rounded perspective and accurate information gathering.

• Discover: "Discover" feature empowers users to grasp intricate concepts and words in online articles. This function ensures a comprehensive understanding of the material being read.

• Ask Anything: Powered by ChatGPT, the "Ask Anything" feature enables users to ask questions and receive easy-to-understand answers, even for the most complex information.

• Summarize: To help users read more efficiently, Wiseone's "Summarize" feature provides the essential information in a condensed format, allowing for quick understanding without missing critical details.

• Suggestions: Wiseone offers a curated list of articles from diverse sources, helping users deepen their understanding of a particular subject.

Probably the next best thing to having GPT-4 connected to the internet.

5 Most Viewed Papers from the past 7 days

arXiv is a free online library where scientists share their research papers before they are published. These are the 5 most viewed papers in the last week.

1. BloombergGPT: A Large Language Model for Finance: A finance-specific Chatbot trained on a large corpus of finance data. See above article!

2. Scalable Multi-Chain Coordination via the Hierarchical Longest Chain Rule: BlockReduce is a system that significantly speeds up digital transactions by organizing them into multiple connected groups, allowing for faster and smoother information sharing between these groups.

3. HuggingGPT: Solving AI Tasks with ChatGPT and its Friends in HuggingFace: HuggingGPT is a system that combines ChatGPT with a variety of AI models to tackle complex tasks across language, vision, and speech, paving a new way towards more advanced and versatile AI.

4. ChatGPT Outperforms Crowd-Workers for Text-Annotation Tasks: ChatGPT is more accurate, consistent, and cheaper than humans at organizing and labeling text, and may disrupt MTurk.

5. LLaMA-Adapter: Efficient Fine-tuning of Language Models with Zero-init Attention: LLaMA-Adapter is a fast and efficient way to improve an AI model's ability to follow instructions and work with images for better problem-solving.

3x the information, for less than $2/week

Stay informed, stay ahead: Your premium AI resource.

Thank you for reading today’s edition.

Your feedback is valuable.
Respond to this email and tell us how you think we could add more value to this newsletter.

Read by employees from