Large Language Models (LLMs) are revolutionizing the development of AI assistants capable of performing diverse tasks across domains. However, current state-of-the-art LLM-driven agents face...

What if your LLM is… a graph? A few days ago, Petar Veličković from Google DeepMind gave one of the most interesting and thought provoking conference I've seen in a while, "Large Language Models as Graph Neural Networks". Once you start seeing LLM as graph neural network, many structural oddities suddenly falls into place. For instance, OpenAI currently recommends to put the instructions at the top of a long prompt. Why is that so? Because due to the geometry of attention graphs, LLM are counter-intuitively biased in favors of the first tokens: they travel constinously through each generation steps, are internally repeated a lot and end up "over-squashing" the latter ones. Models then use a variety of internal metrics/transforms like softmax to moderate this bias and better ponderate distribution, but this is a late patch that cannot solve long time attention deficiencies, even more so for long context. The most interesting aspect of the conference from an applied perspective: graph/geometric representations directly affect accuracy and robustness. As the generated sequence grow and deal with sequences of complex reasoning steps, cannot build solid expert system when attention graphs have single point of failures. Or at least, without extrapolating this information in the first place and providing more detailed accuracy metrics. I do believe LLM explainability research is largely underexploited right now, despite being accordingly a key component of LLM devops in big labs. If anything, this is literal "prompt engineering", seeing models as nearly physical structure under stress and providing the right feedback loops to make them more reliable. | 30 comments on LinkedIn

Choosing the Right Format: How Knowledge Graph Layouts Impact AI Reasoning ... 👉 Why This Matters Most AI systems blend knowledge graphs (structured data) with large language models (flexible reasoning). But there’s a hidden variable: "how" you translate the graph into text for the AI. Researchers discovered that the formatting choice alone can swing performance by up to "17.5%" on reasoning tasks. Imagine solving 1 in 5 more problems correctly just by adjusting how you present data. 👉 What They Built KG-LLM-Bench is a new benchmark to test how language models reason with knowledge graphs. It includes five tasks: - Triple verification (“Does this fact exist?”) - Shortest path finding (“How are two concepts connected?”) - Aggregation (“How many entities meet X condition?”) - Multi-hop reasoning (“Which entities linked to A also have property B?”) - Global analysis (“Which node is most central?”) The team tested seven models (Claude, GPT-4o, Gemini, Llama, Nova) with five ways to “textualize” graphs, from simple edge lists to structured JSON and semantic web formats like RDF Turtle. 👉 Key Insights 1. Format matters more than assumed:   - Structured JSON and edge lists performed best overall, but results varied by task.   - For example, JSON excels at aggregation tasks (data is grouped by entity), while edge lists help identify central nodes (repeated mentions highlight connections). 2. Models don’t cheat: Replacing real entity names with fake ones (e.g., “France” → “Verdania”) caused only a 0.2% performance drop, proving models rely on context, not memorized knowledge. 3. Token efficiency:   - Edge lists used ~2,600 tokens vs. JSON-LD’s ~13,500. Shorter formats free up context space for complex reasoning.   - But concise ≠ always better: structured formats improved accuracy for tasks requiring grouped data. 4. Models struggle with directionality:   Counting outgoing edges (e.g., “Which countries does France border?”) is easier than incoming ones (“Which countries border France?”), likely due to formatting biases. 👉 Practical Takeaways - Optimize for your task: Use JSON for aggregation, edge lists for centrality. - Test your model: The best format depends on the LLM—Claude thrived with RDF Turtle, while Gemini preferred edge lists. - Don’t fear pseudonyms: Masking real names minimally impacts performance, useful for sensitive data. The benchmark is openly available, inviting researchers to add new tasks, graphs, and models. As AI handles larger knowledge bases, choosing the right “data language” becomes as critical as the reasoning logic itself. Paper: [KG-LLM-Bench: A Scalable Benchmark for Evaluating LLM Reasoning on Textualized Knowledge Graphs] Authors: Elan Markowitz, Krupa Galiya, Greg Ver Steeg, Aram Galstyan

Our first attempts at mechanistic interpretability of Transformers from the perspective of network science and graph theory! Check out our preprint: arxiv.org/abs/2502.12352 A wonderful collaboration with superstar MPhil students Batu El, Deepro Choudhury, as well as Pietro Lio' as part of the Geometric Deep Learning class last year at University of Cambridge Department of Computer Science and Technology We were motivated by Demis Hassabis calling AlphaFold and other AI systems for scientific discovery as ‘engineering artifacts’. We need new tools to interpret the underlying mechanisms and advance our scientific understanding. Graph Transformers are a good place to start. The key ideas are: - Attention across multi-heads and layers can be seen as a heterogenous, dynamically evolving graph. - Attention graphs are complex systems represent information flow in Transformers. - We can use network science to extract mechanistic insights from them! More to come on the network science perspective to understanding LLMs next! | 13 comments on LinkedIn

Join our slack and come to the next Graph Learning on Wednesdays (GLOW) session.https://sites.google.com/view/graph-learning-on-wedsOn March 26th, 2025, we h...

LLMs are taking Graph Neural Networks to the next level: While we've been discussing LLMs for natural language, they're quietly changing how we represent…

🎁⏳ Zep packs Temporal Knowledge Graphs + Semantic Entity Extraction + Cypher Queries storage—Outperforming MemGPT with 94.8% Accuracy. Build Personalized AI… | 46 comments on LinkedIn

🏆🚣MiniRAG Introduces Near-LLM Accurate RAG for Small Language Models with Just 25% of the Storage. Achieving that by Semantic-Aware Heterogeneous Graph…

I love Markus J. Buehler's work, and his latest paper "Agentic Deep Graph Reasoning Yields Self-Organizing Knowledge Networks" does not disappoint, revealing… | 19 comments on LinkedIn

🏆🚣MiniRAG Introduces Near-LLM Accurate RAG for Small Language Models with Just 25% of the Storage. Achieving that by Semantic-Aware Heterogeneous Graph…

Breaking LLM Hallucinations in a Smarter Way! (It’s not about feeding more data) Large Language Models (LLMs) still struggle with factual inaccuracies, but…

KET-RAG: Turbocharging AI Agents with 10x Cheaper, Smarter Knowledge Retrieval This Multi-Granular Graph Framework uses PageRank and Keyword-Chunk Graph to have the Best Cost-Quality Tradeoff ﹌﹌﹌﹌﹌﹌﹌﹌﹌ 》The Problem: Knowledge Graphs Are Expensive (and Clunky) AI agents need context to answer complex questions—like connecting “COVID vaccines” to “myocarditis risks” across research papers. But today’s solutions face two nightmares: ✸ Cost: Building detailed knowledge graphs with LLMs can cost $33,000 for a 5GB legal case. ✸ Quality: Cheap methods (like KNN graphs) miss key relationships, leading to 32% worse answers. ☆ Imagine training an AI doctor that either bankrupts you or misdiagnoses patients. Ouch. ﹌﹌﹌﹌﹌﹌﹌﹌﹌ 》The Fix: KET-RAG’s Two-Layer Brain KET-RAG merges precision (knowledge graphs) and efficiency (keyword-text maps) into one system: ✸ Layer 1: Knowledge Graph Skeleton ☆ Uses PageRank to find core text chunks (like “vaccine side effects” in medical docs). ☆ Builds a sparse graph only on these chunks with LLMs—saving 80% of indexing costs. ✸ Layer 2: Keyword-Chunk Bipartite Graph ☆ Links keywords (e.g., “myocarditis”) to all related text snippets—no LLM needed. ☆ Acts as a “fast lane” for retrieving context without expensive entity extraction. ﹌﹌﹌﹌﹌﹌﹌﹌﹌ 》Results: Beating Microsoft’s Graph-RAG with Pennies On HotpotQA and MuSiQue benchmarks, KET-RAG: ✸ Retrieves 81.6% of critical info vs. Microsoft’s 74.6%—with 10x lower cost. ✸ Boosts answer accuracy (F1 score) by 32.4% while cutting indexing bills by 20%. ✸ Scales to terabytes of data without melting budgets. ☆ Think of it as a Tesla Model 3 outperforming a Lamborghini at 1/10th the price. ﹌﹌﹌﹌﹌﹌﹌﹌﹌ 》Why AI Agents Need This AI agents aren’t just chatbots—they’re problem solvers for medicine, law, and customer service. KET-RAG gives them: ✸ Real-time, multi-hop reasoning: Connecting “drug A → gene B → side effect C” in milliseconds. ✸ Cost-effective scalability: Deploying agents across millions of documents without going broke. ✸ Adaptability: Mixing precise knowledge graphs (for critical data) with keyword maps (for speed). Paper in comments ≣≣≣≣≣≣≣≣≣≣≣≣≣≣≣≣≣≣≣≣≣≣≣≣≣≣ 》Build Your Own Supercharged AI Agent? 🔮 Join My 𝐇𝐚𝐧𝐝𝐬-𝐎𝐧 𝐀𝐈 𝐀𝐠𝐞𝐧𝐭𝐬 𝐓𝐫𝐚𝐢𝐧𝐢𝐧𝐠 TODAY! and Learn Building AI Agent with Langgraph/Langchain, CrewAI and OpenAI Swarm + RAG Pipelines 𝐄𝐧𝐫𝐨𝐥𝐥 𝐍𝐎𝐖 [34% discount]: 👉 https://lnkd.in/eGuWr4CH | 10 comments on LinkedIn

Dynamic Reasoning Graphs + LLMs = 🤝 Large Language Models (LLMs) often stumble on complex tasks when confined to linear reasoning. What if they could dynamically restructure their thought process like humans? A new paper introduces Adaptive Graph of Thoughts (AGoT), a test-time framework that replaces rigid prompting strategies (like Chain/Tree of Thought) with dynamic directed acyclic graphs (DAGs). Instead of forcing fixed reasoning steps, AGoT recursively decomposes problems into sub-tasks, selectively expanding only the most critical pathways. This is crucial for industries like scientific research or legal analysis, where problems demand non-linear, nested reasoning. The key innovation lies in complexity checks: AGoT assesses each reasoning node, spawning sub-graphs for intricate subtasks while resolving simpler ones directly. This mirrors how experts allocate mental effort—drilling into uncertainties while streamlining obvious steps. The framework achieved 46.2% improvement on GPQA (a notoriously hard science QA benchmark), rivaling gains from compute-heavy fine-tuning. By unifying chain, tree, and graph paradigms, AGoT retains CoT’s clarity, ToT’s exploration, and GoT’s flexibility without manual tuning. The result? LLMs that self-adapt their reasoning depth based on problem complexity—no architectural changes needed. For AI practitioners, AGoT’s DAG structure offers a principled interface to scale reasoning modularly. ↓ 𝐖𝐚𝐧𝐧𝐚 𝐤𝐧𝐨𝐰 𝐰𝐡𝐚𝐭 𝐲𝐨𝐮 𝐦𝐢𝐬𝐬𝐞𝐝? Join my newsletter with 50k+ readers that breaks down all you need to know about the latest LLM research: llmwatch.com 💡

🚀 Introducing GFM-RAG: The First Graph Foundation Model for Retrieval-Augmented Generation! We’re excited to share our latest research: GFM-RAG: Graph… | 20 comments on LinkedIn

Dynamic Reasoning Graphs + LLMs = 🤝 Large Language Models (LLMs) often stumble on complex tasks when confined to linear reasoning. What if they could… | 10 comments on LinkedIn

The recently developed retrieval-augmented generation (RAG) technology has enabled the efficient construction of domain-specific applications. However, it also has limitations, including the gap...

Terminology Augmented Generation (TAG)? Recently some fellow terminologists have proposed the new term "Terminology-Augmented Generation (TAG)" to refer to… | 29 comments on LinkedIn

What is really Graph RAG? Inspired by "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" paper from Microsoft! How do you combine… | 12 comments on LinkedIn

Knowledge Graphs as a source of trust for LLM-powered enterprise question answering That has been our position from the beginning when we started our research… | 29 comments on LinkedIn

Ashley Faith hosted Amy Hodler and me on a recent episode about “2025 Predictions” — here are notes I prepared, to review in 2026…

This paper presents OG-RAG, an Ontology-Grounded Retrieval Augmented Generation method designed to enhance LLM-generated responses by anchoring retrieval processes in domain-specific ontologies....

Much has been discussed about how Large Language Models, Knowledge Graphs and Search Engines can be combined in a synergistic manner. A dimension largely absent from current academic discourse is...

Task planning in language agents is emerging as an important research topic alongside the development of large language models (LLMs). It aims to break down complex user requests in natural...

SimGRAG is a novel method for knowledge graph driven RAG, transforms queries into graph patterns and aligns them with candidate subgraphs using a graph…

Graphs + Transformers = the best of both worlds 🤝 The same models powering breakthroughs in natural language processing are now being adapted for graphs…

Unlocking universal reasoning across knowledge graphs. Knowledge graphs (KGs) are powerful tools for organizing and reasoning over vast amounts of… | 11 comments on LinkedIn

My takeaways from the International Semantic Web Conference #iswc2024 Ioana keynote: Great example of data integration for journalism, highlighting the use of…

Beyond Vector Space : Knowledge Graphs and the New Frontier of Agentic System Accuracy ⛳ In the realm of agentic systems, a fundamental challenge emerges…

👉 Our paper "Understanding SPARQL Queries: Are We Already There?", explores the potential of Large Language Models (#LLMs) to generate natural-language…

Language agents have recently been used to simulate human behavior and user-item interactions for recommendation systems. However, current language agent simulations do not understand the...