LAI #70: Math Behind Ghibli-Fication, MCP, Deep Research Tools, and Quantum!

Author(s): Towards AI Editorial Team

Originally published on Towards AI.

Good morning, AI enthusiasts! There’s a lot going on in the world of AI lately, and we know it’s not always easy to keep up. So, in this issue, we’ve rounded up some of the most interesting updates and resources — both from the community and our own team — to help you stay in the loop. Enjoy the read!

What’s AI Weekly

I attended Jensen’s Quantum panel with 14 leaders in the field at GTC, and this week, I want to share some bits from it. With all my time spent in AI, quantum has always felt a bit abstract and hyped. The promise is that quantum computing could fundamentally change what’s possible in science and technology. Despite this, quantum computing often seems shrouded in hype and confusion. This panel, however, actually helped me cut through some of that noise. Read my biggest takeaways about the challenges in quantum, or watch the video on YouTube!

— Louis-François Bouchard, Towards AI Co-founder & Head of Community

Learn AI Together Community section!

Featured Guest Post

We have yet another guest post this week, this time with GradientFlow (aka Ben Lorica) and diving into the hottest new development in AI: Deep Research Tools. While tools like ChatGPT’s web browsing or Perplexity extend these capabilities by gathering context from the internet, they remain limited for complex analytical work. Deep research tools change that by combining conversational AI with autonomous web browsing, tool integrations, and sophisticated reasoning capabilities. If you’re building or integrating AI tools, this is essential context for what’s coming next.

We’ll provide a detailed comparative analysis of popular deep research platforms, examining their unique approaches and explaining why they represent a fundamental shift in knowledge work. Read the complete article here!

AI poll of the week!

If you’re a developer building an AI app and want it to be able to initiate tasks in other apps — not just access data — then MCP is well worth a look. If you don’t know how to actually use it yet, tell us in the thread, and we can discuss and share more information!

Collaboration Opportunities

The Learn AI Together Discord community is flooding with collaboration opportunities. If you are excited to dive into applied AI, want a study partner, or even want to find a partner for your passion project, join the collaboration channel! Keep an eye on this section, too — we share cool opportunities every week!

1. Nareshmeena1202 wants to collaborate with the community to explore ML/DL techniques & research papers, work on projects, and participate in Kaggle competitions. If this sounds like your thing, reach out to him in the thread!

2. Lokomatic has launched an AI agency focused on AI literacy and acceleration and is open to partnerships. If you want to know more, connect with him in the thread!

3. Mendzel. wants to form a study group of up to 3 people to participate in Waymo Open Challenges and Kaggle competitions and create a self-driving car project in Webots environment. If you want to join, contact him in the thread!

Meme of the week!

Meme shared by hitoriarchie

TAI Curated Section

Article of the week

Reinforcement Learning-Enhanced Gradient Boosting Machines By Shenggang Li

This article introduces a novel approach to enhancing Gradient Boosting Machines (GBM) by integrating Reinforcement Learning (RL) to dynamically adjust learning rates during training. The RL agent optimizes gradient scaling factors at each iteration, improving predictive accuracy for both regression and classification tasks. Experiments on Kaggle and synthetic datasets demonstrated that RL-GBM outperformed traditional models like XGBoost, LightGBM, and Random Forest in regression tasks, achieving lower errors. While the RL-GBM classifier showed competitive results, it did not consistently surpass advanced models. It also highlights potential improvements for future research, such as advanced RL algorithms and refined reward functions. This piece is ideal for data scientists and machine learning practitioners interested in innovative methods to boost model performance and adapt to complex data patterns.

Our must-read articles

1. Building Local OCR Application SmolDocling: A Step-by-Step Guide [Part 1] & [Part 2] By Youssef Hosni

This two-part guide details the creation of a local OCR application using SmolDocling, a compact 256M vision-language model designed for efficient, end-to-end document conversion. Part 1 explains the OCR pipeline setup, including environment configuration, model initialization, and text extraction from images. Part 2 integrates the pipeline into a Streamlit-based web application, enabling users to upload images, extract text, and download results in Markdown format. The application is optimized for performance, leveraging GPU acceleration and caching. This guide is ideal for developers and AI enthusiasts seeking to build lightweight, privacy-focused OCR solutions without relying on external APIs.

2. The Math Behind Ghibli-Fication. By Surya Maddula

This blog explains the technical foundations behind the “Ghibli-fiction” feature in OpenAI’s ChatGPT, which generates Studio Ghibli-style images from text prompts. It highlights diffusion models, which iteratively add and remove noise to create high-quality images. The process involves mathematical techniques like Gaussian noise addition, denoising, and cross-attention mechanisms to align text prompts with visual outputs. It also discusses GPT-4o’s architecture, which integrates text and image generation into a single neural network, enabling real-time, detailed image creation. This piece is ideal for AI enthusiasts and professionals interested in understanding the math and mechanics behind generative image models.

3. Break The Vector Search Dependency for Truly Robust RAG Systems By Thuwarakesh Murallie

This blog discusses the limitations of relying solely on vector search for Retrieval-Augmented Generation (RAG) systems and advocates for a hybrid approach combining vector and keyword searches. While vector search excels in semantic understanding, it struggles with exact matches for uncommon abbreviations or specific terms. Conversely, keyword-based methods like BM25 handle exact matches well but lack semantic context. It explores hybrid search, which balances both methods using weighted scores, and highlights the role of reranker models to refine results further. Practical implementation using PostgreSQL, Python, and tools like Cohere is detailed, offering a robust framework for improved retrieval. This blog is ideal for AI practitioners, data engineers, and developers looking to enhance RAG systems with more accurate and context-aware retrieval methods.

If you are interested in publishing with Towards AI, check our guidelines and sign up. We will publish your work to our network if it meets our editorial policies and standards.

Join thousands of data leaders on the AI newsletter. Join over 80,000 subscribers and keep up to date with the latest developments in AI. From research to projects and ideas. If you are building an AI startup, an AI-related product, or a service, we invite you to consider becoming a sponsor.

Published via Towards AI