www.artificialintelligenceupdate.com

Tag: General LLM retrieval-augmented generation

Anthropic’s Contextual RAG and Hybrid Search

Imagine an AI that’s not just informative but super-smart, remembering where it learned things! This is Retrieval Augmented Generation (RAG), and Anthropic is leading the charge with a revolutionary approach: contextual retrieval and hybrid search. Forget basic keyword searches – Anthropic’s AI understands the deeper meaning of your questions, providing thoughtful and relevant answers. This paves the way for smarter customer service bots, personalized AI assistants, and powerful educational tools. Dive deeper into the future of AI with this blog post! Contextual RAG

Anthropic’s Contextual Retrieval and Hybrid Search: The Future of AI Enhancement

In the world of Artificial Intelligence (AI), the ability to retrieve and generate information efficiently is crucial. As technology advances, methods like Retrieval Augmented Generation (RAG) are reshaping how we interact with AI. One of the newest players in this field is Anthropic, with its innovative approach to contextual retrieval and hybrid search. In this blog post, we will explore these concepts in detail, making it easy for everyone, including a 12-year-old, to understand this fascinating topic.

Table of Contents

  1. What is Retrieval Augmented Generation (RAG)?
  2. Anthropic’s Approach to RAG
  3. Understanding Hybrid Search Mechanisms
  4. Contextual BM25 and Embeddings Explained
  5. Implementation Example Using LlamaIndex
  6. Performance Advantages of Hybrid Search
  7. Future Implications of Contextual Retrieval
  8. Further Reading and Resources

1. What is Retrieval Augmented Generation (RAG)?

Retrieval Augmented Generation (RAG) is like having a super-smart friend who can not only tell you things but also remembers where the information came from! Imagine when you ask a question; instead of just giving you a general answer, this friend pulls relevant information from books and articles, mixes that with their knowledge, and provides you with an answer that’s spot on and informative.

Why is RAG Important?

The main purpose of RAG is to improve the quality and relevance of the answers generated by AI systems. Traditional AI models might give you good information, but not always the exact answer you need. RAG changes that by ensuring the AI retrieves the most relevant facts before generating its answer. For further details, check out this introduction to RAG.

2. Anthropic’s Approach to RAG

Anthropic, an AI research organization, has developed a new methodology for RAG that is truly groundbreaking. This method leverages two different techniques: traditional keyword-based searches and modern contextual embeddings.

What are Keyword-Based Searches?

Think of keyword-based search as looking for a specific word in a book. If you type "cat" into a search engine, it looks for pages containing the exact word "cat." This traditional method is powerful but can be limited as it doesn’t always understand the context of your question.

What are Contextual Embeddings?

Contextual embeddings are a newer way of understanding words based on their meanings and how they relate to one another. For example, the word "train," in one sentence, can refer to a mode of transport, while in another, it might mean an exercise routine. Contextual embeddings help the model understand these differences.

The Combination

By blending keyword-based searching and contextual embeddings, Anthropic’s approach creates a more robust AI system that understands context and can respond more accurately to user questions. For more on Anthropic’s approach, visit the article here.

3. Understanding Hybrid Search Mechanisms

Hybrid search mechanisms make AI smarter! They combine the strengths of both keyword precision and semantic (meaning-based) understanding.

How Does it Work?

When you search for something, the AI first looks for keywords to get the basic idea. Then, it examines the context to understand your real intent. This allows it to pull out relevant pieces of information and provide a thoughtful answer that matches what you are really asking.

4. Contextual BM25 and Embeddings Explained

BM25 is a famous algorithm used for ranking the relevance of documents based on a given query. Think of it as a librarian who knows exactly how to find the best books for your request.

What is Contextual BM25?

Contextual BM25 takes the original BM25 algorithm and adds a twist: it considers the context of your questions while ranking the search results. This is like a librarian who not only knows the books but understands what kind of story you enjoy most, allowing them to recommend the perfect match for your interests!

How About Contextual Embeddings?

These help the AI recognize the deeper meaning of phrases. So if you type "I love going to the beach," the AI understands that "beach" is associated with summer, sun, and fun. This allows it to provide answers about beach activities rather than just information about sand.

5. Implementation Example Using LlamaIndex

Let’s take a look at how Anthropic’s contextual retrieval works in practice! LlamaIndex is a fantastic tool that provides a step-by-step guide on implementing these concepts.

Example Code Breakdown

Here is a simple code example illustrating how you might implement a contextual retrieval mechanism using LlamaIndex:

from llama_index import ContextualRetriever

# Create a contextual retriever instance
retriever = ContextualRetriever()

# Define your query
query = "What can I do at the beach?"

# Get the results
results = retriever.retrieve(query)

# Display the results
for result in results:
    print(result)

Explanation of the Code

  • Import Statement: This imports the necessary module to implement the contextual retrieval.
  • Creating an Instance: We create an instance of ContextualRetriever, which will help us search for relevant information.
  • Defining a Query: Here, we determine what we want to ask (about the beach).
  • Retrieving Results: The retrieve method of our instance pulls back suitable answers based on our question.
  • Displaying the Results: This loop prints out the results so you can easily read them.

For more detailed guidance, check out the LlamaIndex Contextual Retrieval documentation.

6. Performance Advantages of Hybrid Search

When comparing traditional models to those using hybrid search techniques like Anthropic’s, the results speak volumes!

Why Is It Better?

  1. Accuracy: Hybrid search ensures that the answers are not only correct but also relevant to user queries.
  2. Context Awareness: It captures user intent better, making interactions feel more like human conversation.
  3. Complex Queries: For challenging questions requiring nuance, this methodology excels in providing richer responses.

Real-World Examples

Studies have shown that systems utilizing this hybrid method tend to outperform older models, particularly in tasks requiring detailed knowledge, such as technical support and educational queries.

7. Future Implications of Contextual Retrieval

As technology continues to evolve, methods like Anthropic’s contextual retrieval are expected to lead the way for even more sophisticated AI systems.

Possible Applications

  • Customer Service Bots: These bots can provide detailed, context-aware help, improving customer satisfaction.
  • Educational Tools: They can assist students by delivering nuanced explanations and relevant examples through adaptive learning.
  • Interactive AI Assistants: These assistants can offer personalized and contextually relevant suggestions by understanding queries on a deeper level.

8. Further Reading and Resources

If you want to dive deeper into the world of Retrieval Augmented Generation and hybrid search, check out these articles and resources:

In summary, Anthropic’s contextual retrieval and hybrid search represent a revolutionary step forward in the RAG methodology. By using a combination of traditional search techniques and modern contextual understanding, AI models can now provide more detailed, relevant, and contextually appropriate responses. This mixture ensures AI responses not only answer questions accurately but also resonate well with users’ needs, leading to exciting applications in various fields. The future of AI is bright, and we have much to look forward to with such innovations!

References

  1. How Contextual Retrieval Elevates Your RAG to the Next Level Comments14 ; What are AI Agents? IBM Technology · 526K views ;…
  2. A Brief Introduction to Retrieval Augmented Generation(RAG) The best RAG technique yet? Anthropic’s Contextual Retrieval and Hybrid Search…
  3. Anthropic’s New RAG Approach | Towards AI Hybrid Approach: By combining semantic search with…
  4. Powerful RAG Using Hybrid Search(Keyword+vVector … – YouTube … RAG Using Hybrid Search(Keyword+vVector search…
  5. RAG vs. Long-Context LLMs: A Comprehensive Study with a Cost … The authors propose a hybrid approach, termed #SELF_ROU…
  6. Query Understanding: A Manifesto Anthropic’s Contextual Retrieval and Hybrid Search. How combining …
  7. Hybrid Search for RAG in DuckDB (Reciprocal Rank Fusion) Hybrid Search for RAG in DuckDB (Reciprocal Rank Fusion). 1.1K …..
  8. Top RAG Techniques You Should Know (Wang et al., 2024) Query Classification · Chunking · Metadata & Hybrid Search · Embedding Model ·…
  9. Contextual Retrieval for Enhanced AI Performance – Amity Solutions RAG retrieves relevant information from a knowledge base a…
  10. Contextual Retrieval – LlamaIndex Contextual Retrieval¶. In this notebook we will demonst…

Citation

  1. Scaling RAG from POC to Production | by Anurag Bhagat | Oct, 2024 The best RAG technique yet? Anthropic’s Contextual Ret…
  2. Stop using a single RAG approach – Steve Jones The best RAG technique yet? Anthropic’s Contextual Retrieval and …
  3. Bridging the Gap Between Knowledge and Creativity: An … – Cubed The best RAG technique yet? Anthropic’s Contextual Retr…
  4. Understanding Vectors and Building a RAG Chatbot with Azure … The best RAG technique yet? Anthropic’s Contextual…
  5. Copilot: RAG Made Easy? – ML6 blog The best RAG technique yet? Anthropic’s Contextual Ret…
  6. Building Smarter Agents using LlamaIndex Agents and Qdrant’s … The best RAG technique yet? Anthropic’s Contextual Retrieval and Hybrid Se…
  7. Building with Palantir AIP: Logic Tools for RAG/OAG The best RAG technique yet? Anthropic’s Contextual Retrieval and Hybri…
  8. Advanced RAG 03 – Hybrid Search BM25 & Ensembles – YouTube The Best RAG Technique Yet? Anthropic’s Contextual…
  9. Anthropic Claude3— a competetive perspective for OpenAI’s GPT … The best RAG technique yet? Anthropic’s Contextual Retriev…

  10. Advanced RAG Techniques: an Illustrated Overview | by IVAN ILIN A comprehensive study of the advanced retrieval augment…

    Don’t miss out on future content—follow us on LinkedIn for the latest updates. Contextual RAG

    Continue your AI exploration—visit AI&U for more insights now.

Retrieval Augmented Generation: RAGatouille

This engaging excerpt dives into RAGatouille, a groundbreaking open-source project that simplifies building powerful AI systems. It combines information retrieval and generation, allowing you to create applications that answer questions, retrieve documents, and even generate content – all efficiently and accurately.

Ready to explore the exciting world of Retrieval-Augmented Generation? Dive into the full guide and unlock the potential of AI for your projects!

RAGatouille: A Comprehensive Guide to Retrieval-Augmented Generation Models

Introduction

In the rapidly evolving world of artificial intelligence and natural language processing (NLP), the ability to retrieve and generate information efficiently is paramount. One of the exciting advancements in this field is the concept of Retrieval-Augmented Generation (RAG). At the forefront of this innovation is RAGatouille, an open-source project developed by AnswerDotAI. This blog post will delve deep into RAGatouille, exploring its features, usage, and the potential it holds for developers and researchers alike.

What is RAGatouille?

RAGatouille is a user-friendly framework designed to facilitate the integration and training of RAG models. By combining retrieval mechanisms with generative models, RAGatouille allows users to create sophisticated systems capable of answering questions and retrieving relevant documents from large datasets.

Key Features of RAGatouille

  1. Ease of Use: RAGatouille is designed with simplicity in mind. Users can quickly set up and start training models without needing extensive configuration or prior knowledge of machine learning.

  2. Integration with LangChain: As a retriever within the LangChain framework, RAGatouille enhances the versatility of applications built with language models. This integration allows developers to leverage RAGatouille’s capabilities seamlessly.

  3. Fine-tuning Capabilities: The library supports the fine-tuning of models, enabling users to adapt pre-trained models to specific datasets or tasks. This feature is crucial for improving model performance on niche applications.

  4. Multiple Examples and Notebooks: RAGatouille comes with a repository of Jupyter notebooks that showcase various functionalities, including basic training and fine-tuning without annotations. You can explore these examples in the RAGatouille GitHub repository.

  5. Community Engagement: The active GitHub repository fosters community involvement, allowing users to report issues, ask questions, and contribute to the project. Engaging with the community is essential for troubleshooting and learning from others’ experiences.

Getting Started with RAGatouille

Installation

Before diving into the functionalities of RAGatouille, you need to install it. You can do this using pip:

pip install ragatouille

Basic Usage

Let’s start with a simple code example that demonstrates the basic usage of RAGatouille for training a model.

from ragatouille import RAGTrainer
from ragatouille.data import DataLoader

# Initialize the trainer
trainer = RAGTrainer(model_name="MyFineTunedColBERT", pretrained_model_name="colbert-ir/colbertv2.0")

# Load your dataset
data_loader = DataLoader("path/to/your/dataset")

# Train the model
trainer.train(data_loader)

Breakdown of the Code:

  1. Importing Modules: We import the necessary classes from the RAGatouille library.
  2. Initializing the Trainer: We create an instance of RAGTrainer, specifying the model we want to fine-tune.
  3. Loading the Dataset: We load our dataset using the DataLoader class.
  4. Training the Model: Finally, we call the train method to begin the training process.

This straightforward approach allows users to set up a training pipeline quickly.

Fine-Tuning a Model

Fine-tuning is essential for adapting pre-trained models to specific tasks. RAGatouille provides a simple way to fine-tune models without requiring annotated data. Here’s an example of how to do this:

from ragatouille import RAGFineTuner
from ragatouille.data import DataLoader

# Initialize the fine-tuner
fine_tuner = RAGFineTuner(model_name="MyFineTunedModel", pretrained_model_name="colbert-ir/colbertv2.0")

# Load your dataset
data_loader = DataLoader("path/to/your/dataset")

# Fine-tune the model
fine_tuner.fine_tune(data_loader)

Understanding the Fine-Tuning Process

  1. Fine-Tuner Initialization: We create an instance of RAGFineTuner with a specified model.
  2. Loading the Dataset: The dataset is loaded similarly to the training example.
  3. Fine-Tuning the Model: The fine_tune method is called to adapt the model to the dataset.

This flexibility allows developers to enhance model performance tailored to their specific needs.

Advanced Features

Integration with LangChain

LangChain is a powerful framework for developing applications that utilize language models. RAGatouille’s integration with LangChain allows users to harness the capabilities of both tools effectively. This integration enables developers to build applications that can retrieve information and generate text based on user queries.

Community and Support

RAGatouille boasts an active community on GitHub, where users can report bugs, seek help, and collaborate on features. Engaging with the community is crucial for troubleshooting and learning from others’ experiences.

Use Cases for RAGatouille

RAGatouille can be applied in various domains, including:

  1. Question-Answering Systems: Organizations can implement RAGatouille to build systems that provide accurate answers to user queries by retrieving relevant documents.

  2. Document Retrieval: RAGatouille can be used to create applications that search large datasets for specific information, making it valuable for research and data analysis.

  3. Chatbots: Developers can integrate RAGatouille into chatbots to enhance their ability to understand and respond to user inquiries.

  4. Content Generation: By combining retrieval and generation, RAGatouille can assist in creating informative content based on user requests.

Interesting Facts about RAGatouille

  • The name "RAGatouille" is a clever play on words, combining Retrieval-Augmented Generation with a nod to the French dish ratatouille, symbolizing the blending of various machine learning elements into a cohesive framework.
  • The project has gained traction on social media and various forums, showcasing its growing popularity and the community’s interest in its capabilities.

Conclusion

RAGatouille stands out as a powerful and user-friendly tool for anyone looking to implement retrieval-augmented generation models efficiently. Its ease of use, robust features, and active community involvement make it an invaluable resource for researchers and developers in the NLP field. Whether you’re building a question-answering system, a document retrieval application, or enhancing a chatbot, RAGatouille provides the tools and support to bring your ideas to life.

Important Links

In summary, RAGatouille is not just a framework; it is a gateway to harnessing the power of advanced NLP techniques, making it accessible for developers and researchers alike. Start exploring RAGatouille today, and unlock the potential of retrieval-augmented generation for your applications!

References

  1. RAGatouille/examples/02-basic_training.ipynb at main – GitHub … RAGatouille/examples/02-basic_training.ipynb at ma…
  2. Question: How to get score of ranked document? · Issue #201 – GitHub Hey all, I’m using RAGatouille as a retriever for lang…
  3. Benjamin Clavié (@bclavie) / X … linearly on a daily basis @answerdotai | cooking some late interaction …
  4. ragatouille | PyPI | Open Source Insights Links. Origin. https://pypi.org/project/ragatouille/0.0.8.post4/. Repo. htt…
  5. Idea: Make CorpusProcessor (and splitter_fn / preprocessing_fn) to … AnswerDotAI / RAGatouille Public. Sponsor · Notifications You must be … …
  6. Compatibility with LangChain 0.2.0 · Issue #215 – GitHub I would like to use ragatouille with langchain 0.2…
  7. Use base model or sentence transformer · Issue #225 – GitHub AnswerDotAI / RAGatouille Public. Sponsor · Notifications You must be …
  8. Steren on X: "After "Mistral", "RAGatouille" by @bclavie https://t.co … https://github.com/bclavie/RAGatouille… Yes to more Fr…
  9. Byaldi: A ColPali-Powered RAGatouille’s Mini Sister Project by … Byaldi: A ColPali-Powered RAGatouille’s Mini Sister Project …..
  10. About Fine-Tuning · Issue #212 · AnswerDotAI/RAGatouille – GitHub I have a few more questions. I would be happy if you answer….
  11. Best opensource rag with ui – Reddit https://github.com/infiniflow/ragflow Take a look at RAGFlow, aiming ….
  12. Question: rerank does not use index · Issue #235 – GitHub AnswerDotAI / RAGatouille Public. Sponsor · Notifications You must be … S…

For more tips and strategies, connect with us on LinkedIn now.

Looking for more AI insights? Visit AI&U now.

Boost LLM’s RAG Performance with Chunking!

Boost AI Performance with Chunking!

This powerful technique breaks down complex information for AI, leading to smarter responses in chatbots, question-answering systems, and more. Discover how chunking unlocks the true potential of RAG architectures.

Dive in and unlock the future of AI!

The Art of Chunking: Boosting AI Performance in RAG Architectures

Introduction

In the rapidly evolving landscape of artificial intelligence (AI), the efficiency and effectiveness of information processing are paramount. One cognitive strategy that has gained attention for its potential to enhance AI performance is chunking—a method that involves breaking down information into smaller, more manageable units or ‘chunks.’ This technique is particularly significant in the context of Retrieval-Augmented Generation (RAG) architectures. RAG combines the strengths of retrieval-based systems with generative models, enabling AI to efficiently handle vast datasets while improving response accuracy and contextual relevance.

In this blog post, we will delve into the intricacies of chunking and its profound impact on enhancing AI performance, especially within RAG architectures. We will explore key concepts, implementation strategies, challenges, and real-world applications, providing a comprehensive understanding of how chunking serves as a critical tool in the AI arsenal.

Understanding RAG Architectures

At the core of RAG architectures lies a dual mechanism that consists of two primary components:

  1. Retriever: This component is responsible for fetching relevant information from a knowledge base. It identifies and retrieves specific data points that are pertinent to a given query, effectively narrowing down the vast sea of information available.

  2. Generator: Once the retriever has fetched the relevant information, the generator constructs coherent and contextually appropriate responses based on this data. This generative aspect ensures that the AI can articulate responses that are not only accurate but also fluent and engaging.

The synergy between these components allows RAG systems to leverage extensive datasets while maintaining contextual relevance and coherence in their outputs. However, the effectiveness of this architecture hinges on the ability to process information efficiently—an area where chunking plays a crucial role.

The Role of Chunking in RAG

Chunking simplifies the input data for both the retriever and generator components of RAG systems. By dividing extensive datasets into smaller, contextually relevant segments, AI models can better understand and process information. This method aids in reducing cognitive load, thereby enhancing the model’s ability to generate accurate and context-aware outputs.

Cognitive Load Reduction

Cognitive load refers to the amount of mental effort being used in working memory. In the context of AI, reducing cognitive load can lead to improved performance. When information is chunked into smaller segments, it becomes easier for the AI to process and retrieve relevant data. This is akin to how humans naturally group information—such as remembering a phone number by breaking it down into smaller parts (Sweller, 1988).

Enhanced Contextual Understanding

Chunking also enhances the AI’s ability to maintain context. By organizing information into logical segments, the retriever can more effectively match queries with relevant pieces of information. Similarly, the generator can focus on smaller sets of data, which allows for more precise and relevant output generation.

Performance Improvement

Research indicates that chunking can significantly enhance the retrieval accuracy of RAG systems. When data is broken into logical segments, the retriever can more effectively match queries with relevant pieces of information. This boost in accuracy translates to more reliable AI outputs (Karpukhin et al., 2020).

Empirical Evidence

Studies have shown that RAG architectures that implement chunking demonstrate improved performance metrics. For instance, retrieval accuracy can see marked improvements when the input data is appropriately chunked. Additionally, generative models benefit from chunking as they can concentrate on smaller, meaningful datasets, resulting in outputs that are not only accurate but also contextually rich (Lewis et al., 2020).

Implementation Strategies for RAG

To maximize the benefits of chunking, several effective strategies can be employed:

  1. Semantic Chunking: This involves organizing data based on meaning and context. By grouping information that shares a common theme or subject, AI systems can retrieve and generate more coherent responses.

  2. Structural Chunking: Here, information is grouped according to its format, such as paragraphs, bullet points, or sections. This method allows the AI to recognize patterns in the data, facilitating better retrieval and generation.

  3. Hierarchical Chunking: This strategy organizes information from general to specific. By structuring data in a hierarchy, AI systems can efficiently navigate through layers of information, enhancing retrieval efficiency.

Balancing Chunk Size

While chunking offers numerous benefits, it is essential to balance the size of the chunks. Overly small chunks may lead to a loss of context, making it challenging for the AI to generate coherent responses. Conversely, excessively large chunks might overwhelm the retrieval process, negating the benefits of chunking altogether. Therefore, designing chunking strategies should consider the nature of the data and the specific application of the RAG architecture.

Challenges and Considerations for RAG

Despite its advantages, implementing chunking in RAG architectures comes with challenges. Here are a few considerations:

  1. Context Preservation: Maintaining context while chunking is critical. Developers must ensure that the chunks retain enough information for the AI to understand the overall narrative or argument being presented.

  2. Data Nature: The type of data being processed can influence chunking strategies. For example, textual data may require different chunking methods compared to structured data like spreadsheets.

  3. Real-time Processing: In applications that require real-time responses, such as chatbots, the chunking process must be efficient and rapid to avoid delays in response time.

  4. Adaptability: As AI continues to evolve, chunking strategies must adapt to new types of data and changing user expectations. Continuous evaluation and refinement of chunking methods will be necessary to keep pace with advancements in AI technology.

Applications of Chunking in RAG

Chunking has far-reaching implications in various applications of RAG architectures, particularly in natural language processing (NLP) and information retrieval systems.

Question-Answering Systems

In NLP, chunking can significantly enhance the performance of question-answering systems. By ensuring that the AI retrieves and generates contextually relevant information effectively, users receive accurate and meaningful answers quickly (Chen et al., 2017).

Chatbots and Conversational Agents

For chatbots and conversational agents, chunking enables these systems to maintain context throughout a dialogue. By breaking down user queries and responses into manageable chunks, these AI systems can provide more relevant and coherent interactions, improving user satisfaction.

Document Retrieval Systems

In document retrieval systems, chunking allows for more efficient indexing and searching. By organizing documents into coherent chunks, the retrieval process becomes faster and more accurate, leading to improved user experiences. Users can find the information they need more quickly, enhancing the overall efficiency of the system (Manning et al., 2008).

Conclusion

The art of chunking is an essential technique for enhancing AI performance in Retrieval-Augmented Generation architectures. By breaking down complex information into manageable pieces, chunking not only supports more effective retrieval and generation processes but also improves the overall accuracy and relevance of AI outputs.

As AI continues to evolve, the integration of chunking strategies will play a crucial role in optimizing performance and user interaction across various applications. This comprehensive overview highlights the importance of chunking in boosting AI performance, particularly within RAG architectures, providing valuable insights for researchers, developers, and practitioners in the field.

In conclusion, understanding and implementing chunking strategies can significantly enhance the capabilities of AI systems, ultimately leading to more intelligent and responsive applications that can better serve user needs. The future of AI will undoubtedly benefit from the continued exploration and application of chunking techniques, paving the way for more sophisticated and efficient technologies.

References

  1. Sweller, J. (1988). Cognitive load during problem-solving: Effects on learning. Cognitive Science.
  2. Karpukhin, V., Oguz, B., Min, S., Wu, L., Edunov, S., Chen, D., & Yih, W. (2020). Dense Passage Retrieval for Open-Domain Question Answering. arXiv.
  3. Lewis, P., Perez, E., Piktus, A., Petroni, F., Karpukhin, V., Goyal, N., … & Riedel, S. (2020). Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks. arXiv.
  4. Chen, D., Fisch, A., Weston, J., & Bordes, A. (2017). Reading Wikipedia to Answer Open-Domain Questions. arXiv.
  5. Manning, C. D., Raghavan, P., & Schütze, H. (2008). Introduction to Information Retrieval. Stanford NLP.

Stay ahead in your industry—connect with us on LinkedIn for more insights.

Dive deeper into AI trends with AI&U—check out our website today.

Exit mobile version