LAI #75: Generative AI vs. Agentic AI vs. AI Agents
LAI #75: Generative AI vs. Agentic AI vs. AI Agents
0 like
May 15, 2025
Share this post
Author: Towards AI Editorial Team
Originally published on Towards AI.
Good morning, AI enthusiasts,
This week’s issue dives into where the field is heading — beyond generation, toward autonomy and better error awareness. We’re starting with a breakdown of the increasingly fuzzy but important distinctions between Generative AI, Agentic AI, and AI Agents. Then we move into applied innovation: Microsoft’s GraphRAG, multimodal RAG systems using Cohere and Gemini, and a practical framework for predicting when your model is about to get something wrong.
Also in the mix: DNNs vs. tree-based models for e-commerce ranking, a powerful Cursor.ai-like browser extension from the community, and this week’s poll on when vibes are enough — and when accuracy has to come first.
Let’s get into it.
— Louis-François Bouchard, Towards AI Co-founder & Head of Community
Learn AI Together Community Section!
Featured Community post from the Discord
Retconned has built Sophon, an AI chat app that enhances your browsing experience by understanding and interacting with your tabs. With its intelligent composer, it can see the tabs you have open, allowing it to understand context and autofill forms, textboxes, or fields with a single click. It is a browser extension and completely free. Check it out here. Share your feedback in the thread and support a fellow community member!
AI poll of the week!
Most of you are doing vibe checks, and of course, for general tasks, the entire idea is for the AI to not feel like AI. But would you also rely on “vibes” for more quantitative tasks, where output accuracy matters more than output feel? Share in the thread, let’s decide together!
Meme of the week!
Meme shared by rucha8062
TAI Curated Section
Article of the week
How GraphRAG Works Step-by-Step By Mariana Avelino
This blog explains Microsoft’s GraphRAG, a method that uses knowledge graphs for retrieval-augmented generation. Key detailed processes were graph creation, involving entity extraction, community partitioning, and querying, with distinct Local and Global Search functions. It outlined how entities, relationships, and community reports are generated and used for LLM response generation, including context management and semantic retrieval.
Our must-read articles
1. Distill-then-Detect: A Practical Framework for Error-Aware Machine Learning By Shenggang Li
The author presented a framework, “Distill-then-Detect,” to address prediction errors in machine learning models, particularly the “big misses” on critical data slices. This approach involves distilling a compact “student” model from a larger “teacher” model. It then quantifies teacher uncertainty and trains a meta-model to predict where the teacher is likely to err. By combining these signals into a risk score and applying conformal calibration for thresholding, the system effectively flags high-risk predictions. Experiments demonstrated that this method identified error-prone cases with balanced precision and recall while clustering these errors provided actionable insights into problematic data segments.
2. Beyond Text: Building Multimodal RAG Systems with Cohere and Gemini By Sridhar Sampath
Traditional Retrieval-Augmented Generationsystems often fail to process visual data. This article details a multimodal RAG system designed to overcome this limitation by understanding both text and images within documents. It utilizes Cohere’s multimodal embeddings to create unified vector representations from content like PDFs. Gemini 2.5 Flash then generates context-aware answers using either matched text or images, with FAISS managing vector indexing. It explains the system’s workflow, from document upload to answer generation, demonstrating its enhanced capability to extract information from charts, tables, and other visuals compared to text-only RAG.
3. Generative AI vs. Agentic AI vs. AI Agents: What Everyone Needs to Know By Poojan Vig
The article clarified the distinct roles of Generative AI, Agentic AI, and AI Agents. It explained that Generative AI produces new content based on learned patterns. Agentic AI focuses on strategy, planning, and iteration towards a goal without continuous human intervention. AI Agents then sense their environment and execute actions in the digital or real world. Using a cooking analogy and examples like automated customer service, the piece illustrated how these AI types can operate independently or collaboratively to perform complex tasks.
4. DNNs vs Traditional Tree-Based Models for E-Commerce Ranking By Nikhilesh Pandey
The author discusses the evolution of e-commerce ad ranking systems, detailing the shift from traditional tree-based models to Deep Neural Networks. It outlines why tree-based models have reached their limits and how DNNs offer superior capabilities for handling complex data, personalization, and achieving better conversion ratepredictions. Using DoorDash Ads as a case study, the piece illustrates the iterative migration process, including defining baselines, optimizing model training and evaluation with techniques like data normalization and distributed processing, and addressing challenges such as the offline-online performance gap.
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
Towards AI - Medium
Share this post
#lai #generative #agentic #agents
LAI #75: Generative AI vs. Agentic AI vs. AI Agents
LAI #75: Generative AI vs. Agentic AI vs. AI Agents
0 like
May 15, 2025
Share this post
Author: Towards AI Editorial Team
Originally published on Towards AI.
Good morning, AI enthusiasts,
This week’s issue dives into where the field is heading — beyond generation, toward autonomy and better error awareness. We’re starting with a breakdown of the increasingly fuzzy but important distinctions between Generative AI, Agentic AI, and AI Agents. Then we move into applied innovation: Microsoft’s GraphRAG, multimodal RAG systems using Cohere and Gemini, and a practical framework for predicting when your model is about to get something wrong.
Also in the mix: DNNs vs. tree-based models for e-commerce ranking, a powerful Cursor.ai-like browser extension from the community, and this week’s poll on when vibes are enough — and when accuracy has to come first.
Let’s get into it.
— Louis-François Bouchard, Towards AI Co-founder & Head of Community
Learn AI Together Community Section!
Featured Community post from the Discord
Retconned has built Sophon, an AI chat app that enhances your browsing experience by understanding and interacting with your tabs. With its intelligent composer, it can see the tabs you have open, allowing it to understand context and autofill forms, textboxes, or fields with a single click. It is a browser extension and completely free. Check it out here. Share your feedback in the thread and support a fellow community member!
AI poll of the week!
Most of you are doing vibe checks, and of course, for general tasks, the entire idea is for the AI to not feel like AI. But would you also rely on “vibes” for more quantitative tasks, where output accuracy matters more than output feel? Share in the thread, let’s decide together!
Meme of the week!
Meme shared by rucha8062
TAI Curated Section
Article of the week
How GraphRAG Works Step-by-Step By Mariana Avelino
This blog explains Microsoft’s GraphRAG, a method that uses knowledge graphs for retrieval-augmented generation. Key detailed processes were graph creation, involving entity extraction, community partitioning, and querying, with distinct Local and Global Search functions. It outlined how entities, relationships, and community reports are generated and used for LLM response generation, including context management and semantic retrieval.
Our must-read articles
1. Distill-then-Detect: A Practical Framework for Error-Aware Machine Learning By Shenggang Li
The author presented a framework, “Distill-then-Detect,” to address prediction errors in machine learning models, particularly the “big misses” on critical data slices. This approach involves distilling a compact “student” model from a larger “teacher” model. It then quantifies teacher uncertainty and trains a meta-model to predict where the teacher is likely to err. By combining these signals into a risk score and applying conformal calibration for thresholding, the system effectively flags high-risk predictions. Experiments demonstrated that this method identified error-prone cases with balanced precision and recall while clustering these errors provided actionable insights into problematic data segments.
2. Beyond Text: Building Multimodal RAG Systems with Cohere and Gemini By Sridhar Sampath
Traditional Retrieval-Augmented Generationsystems often fail to process visual data. This article details a multimodal RAG system designed to overcome this limitation by understanding both text and images within documents. It utilizes Cohere’s multimodal embeddings to create unified vector representations from content like PDFs. Gemini 2.5 Flash then generates context-aware answers using either matched text or images, with FAISS managing vector indexing. It explains the system’s workflow, from document upload to answer generation, demonstrating its enhanced capability to extract information from charts, tables, and other visuals compared to text-only RAG.
3. Generative AI vs. Agentic AI vs. AI Agents: What Everyone Needs to Know By Poojan Vig
The article clarified the distinct roles of Generative AI, Agentic AI, and AI Agents. It explained that Generative AI produces new content based on learned patterns. Agentic AI focuses on strategy, planning, and iteration towards a goal without continuous human intervention. AI Agents then sense their environment and execute actions in the digital or real world. Using a cooking analogy and examples like automated customer service, the piece illustrated how these AI types can operate independently or collaboratively to perform complex tasks.
4. DNNs vs Traditional Tree-Based Models for E-Commerce Ranking By Nikhilesh Pandey
The author discusses the evolution of e-commerce ad ranking systems, detailing the shift from traditional tree-based models to Deep Neural Networks. It outlines why tree-based models have reached their limits and how DNNs offer superior capabilities for handling complex data, personalization, and achieving better conversion ratepredictions. Using DoorDash Ads as a case study, the piece illustrates the iterative migration process, including defining baselines, optimizing model training and evaluation with techniques like data normalization and distributed processing, and addressing challenges such as the offline-online performance gap.
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
Towards AI - Medium
Share this post
#lai #generative #agentic #agents
·33 Vue