Kjell Carlsson, Ph.D.

NEW in the #KempnerInstitute blog: A method to predict how #LLMs scale w/ compute across different datasets. Read it here: https://buff.ly/4iqPw9o

8

Improved coding of PD, LGD and EAD models is one of the potential benefits of this innovative technology. How does this work, and what else can modelers expect from generative AI systems like ChatGPT? Revisit our recent #RiskIntelligence piece on how generative AI may impact credit risk modeling. https://bit.ly/44AMj0m #chatGPT #riskmodeling #AI #generativeAI #financialrisk

115

3 Comments

Getting the Most Value from LLMs Using a Knowledge Architecture, Ontologies, and Knowledge Graphs: Practices and Approaches Highlights ⤵️ This workshop focused on using Large Language Models to boost productivity and enterprise efficiency. A case study from a pharmaceutical company shows how structured information architecture and metadata-enriched prompts increased model accuracy from 53% to 83%. For more insight read Seth's article, https://hubs.ly/Q02wRfVG0. #BioITWorld #LLM #KnowledgeArchitecture #IA @SethEarley @Bio-ITWorld

4

Heading to Databricks's #DataAISummit? Connect with fellow attendees interested in improving LLM evaluations and feedback techniques by grabbing the last available spot for Arjun Bansal's braindate at the event ⬇️ https://lnkd.in/gAT6PTYt If you didn't get a spot in the group, you can still learn about Log10's approach in Arjun's conference session, AutoFeedback: Scaling Human Feedback with Custom Evaluation Models, or at Log10.io #LLMOps #LLMs

9

Are you interested in vector search, indexing, or database efficiency? Then you won't want to miss this talk by Etienne Dilocker from Weaviate at Carnegie Mellon University's Database Group with Andy Pavlo! Etienne breaks down key challenges and innovations that are powering the future of search: ✨ Bit-Sliced Indexing - tackling complex range queries like price filters with innovative indexing. ✨ BM25 Hybrid Search - combining keyword and vector search to handle nuanced queries. ✨ Graph-Based Indexing (HNSW) - optimizing for high update rates and recalls without latency drops. ✨ Overcoming the Minority Filter Problem - enhancing search relevance through in-place filtering, based on the ACORN paper. ✨ Weaviate's Roadmap - performance, enterprise readiness, and pioneering "Recommendations as a Service." Check out the full talk for an in-depth look at how Weaviate is setting new standards in search technology: https://lnkd.in/gavAZRrn

33

2 Comments

Join Carnegie Mellon University's Philip Koopman on Sept 11, 10 AM EDT, for insights on leveraging #AI effectively. Learn about: • Classification & generative outputs • Addressing bias & safety • The 90/10 principle • Real-world Machine Learning applications Understand AI's potential and limitations. Register now! https://lnkd.in/gvhVk33y #MachineLearning #TechInnovation #BOSS2024 #ArtificialIntelligence #DeepLearning #Semiconductor #ML

8

Machine learning helps firms navigate complex datasets, but the opacity of these models can jeopardize regulatory compliance. That makes it vital to ensure transparency. How can this be done while still benefitting from ML predictions? Find answers in our latest blog, written by Davide Bonamico, Alessandro Amato, and Serena Manti. https://lnkd.in/esKcZhA8

6

🏛️ "In many enterprise use cases, especially in highly regulated sectors like government, healthcare, and finance, precision is non-negotiable, often ruling out probabilistic methods like vector search-enabled RAG. In these cases, knowledge graphs come to the rescue, providing the determinism required for reliable data retrieval. Enterprise knowledge graphs are built on carefully crafted taxonomies and ontologies that systematically define and disambiguate concepts, capturing the relationships and hierarchies within the data." Hear from David Hannibal, Chief Product Officer & Head of Corporate Development at Squirro, on how Knowledge Graphs Supercharge Vector Search for RAG ⚡ in our latest article: https://lnkd.in/eWCHiyam

22

Dear friends! We’re happy to share that Pavlin Poličar from our team got his openTSNE paper published in the Journal of Statistical Software 🙌 t-distributed stochastic neighbor embedding (t-SNE) is one of the most popular dimensionality reduction techniques, which are used for making sense of high-dimensional datasets. These datasets contain large numbers of rows and columns, which often makes it difficult to identify any interesting patterns. 🤷‍♂️ But with dimensionality reduction techniques, we can generate two-dimensional embeddings, which can be visualised in a scatter plot. ✍ Nearly six years ago, when Pavlin released openTSNE and put out his initial preprint, existing Python implementations of t-SNE were either prohibitively slow, difficult to set up, or lacked newly researched features. 🙈 For this reason, he set out to write a better implementation - one that would be: ✅ fast, ✅ simple to play around with, ✅ easy to install, and that would ✅ incorporate all the latest developments to generate embeddings as informative as possible. And thus came into existence openTSNE, our new Python implementation of the t-SNE algorithm! In Orange, the t-SNE widget has been powered by openTSNE for the past several years. 🔝 🍊 We wrote about it on our blog 👉 https://lnkd.in/dczq9Nw9 You can read the full paper here 👉 https://lnkd.in/dTb-NXwm To learn more, check out our GitHub repository here 👉 https://lnkd.in/dthjxtm Congratulations, Pavlin! #datascience #dataanalysis #machinelearning #innovation

42

2 Comments

🗞 Read two new blog posts by Joe Massey and Elena Simperl from Open Data Institute, and Vinay Narayan from Aapti! 1️⃣ How can we empower data communities in the era of generative AI? This blog looks at how generative artificial intelligence (GenAI) is altering the ways communities function online! 🔗 https://lnkd.in/gJbZ7JN3 2️⃣ Why the UN Global Digital Compact needs to get serious about participatory data This blog explores the potential of the Global Digital Compact and seeks to unpack how participation in the data economy is an important part of achieving its aims. 🔗 https://lnkd.in/g8wa9ftH #tech #globaldigitalcompact #data #participation

21

🤖Interactions Between Artificial Intelligence and Digital Public Infrastructure IIPP's David Eaves & Sarosh Nagar discuss the opportunities and challenges under which #AI and #DPI can interact for mutual benefit in their latest paper for arXiv. ✍️Read it here: https://lnkd.in/gnw7wFBc

27

1 Comment

#LangTest makes it far easier to test your custom #LLM for #ResponsibleAI against existing models and leaderboards, across multiple benchmarks, right from the command line: https://hubs.li/Q02wJqDz0 #ai #generativeai #largelanguagemodels #ethicalai #modeltesting #modelevaluation #datascience

1

Fine-Tuning Large Language Models to Enhance Entity Matching and Generalization 📌 This paper explores the potential of fine-tuning large language models (LLMs) to enhance entity matching tasks, an essential component in data integration pipelines. While previous research emphasized prompt engineering and in-context learning, this study focuses on fine-tuning LLMs to improve performance, particularly regarding generalization across different datasets and domains. Article link: https://lnkd.in/gRCthTUV 🔹 Entity matching identifies whether different data descriptions refer to the same entity across various sources. Prior studies showed that LLMs such as GPT and Llama outperform PLMs in zero-shot settings. The goal is to enhance their ability through fine-tuning, addressing issues with cross-domain generalization and in-context learning. 🔹 This study introduces two key dimensions: augmenting training examples with explanations and using LLMs to select or generate better training examples. By fine-tuning smaller and larger models, the research reveals how structured explanations and filtered or generated training examples influence performance. This approach contributes a detailed exploration of the potential for fine-tuning to go beyond prompt engineering, particularly in improving generalization within and across domains. 🔹 Fine-tuning leads to notable improvements in smaller models and enhances in-domain generalization. The addition of structured explanations boosts performance across three of four models. However, fine-tuning’s effect on larger models, such as GPT-4o, remains mixed, and cross-domain generalization often suffers. 🔹 Fine-tuning enhances smaller models' performance, especially in specific domains, but struggles to improve cross-domain capabilities. Future research should focus on refining example generation methods to address these cross-domain limitations and improve the generalization capabilities of LLMs. 📑 Steiner, A., Peeters, R., Bizer, C. (2024). Fine-tuning Large Language Models for Entity Matching. DOI: 10.48550/arXiv.2409.08185.

1

🌟Check out this insightful article on @DZoneInc about the future of LLMs. A must-read for anyone interested in the evolving landscape of #AI and #datascience! #LLM #bigdata https://lnkd.in/dVCVxvDK

2

#B2BSaaS operators take note! While just about every data professional on the planet is actively building gen-AI applications for their company, two-thirds do not fully trust the underlying data. Ouch! Barr Moses from Monte Carlo Data argues that this mistrust in data, in tandem with substantial pressure to implement GenAI without realistic expectations, presents a serious risk for businesses. What to do? Set up routine quality data checks. Automate monitoring. Insist on feedback loops between data teams and business users. And when you report on your Gen AI initiatives, highlight the status of data quality along with successes and challenges. https://lnkd.in/d9nMeHye

5

1 Comment

In the rapidly evolving world of LLMs, how do we ensure that models like GPT, LLaMA, and others are truly effective? LLM Benchmarks are the yardstick for evaluating model performance, capabilities, and reliability. Why Are Benchmarks Crucial? - They standardize evaluation across diverse models - Help identify strengths and weaknesses in language comprehension, reasoning, and generation - Aid in making informed decisions about deploying LLMs in real-world applications LLM Benchmarks are more than just numbers - they tell us how well an LLM can adapt, learn, and deliver across different scenarios. Learn more now! https://hubs.la/Q030sc2_0 #LLM #Benchmarks #LLMEvaluation

9

1 Comment

AI data infrastructures are increasingly reliant on advanced storage solutions to manage the sheer volume and velocity of data processing demands, as highlighted in Stephen Foskett's article on Gestalt IT. Solidigm's contributions to the field, with their state-of-the-art storage technologies, underscore the necessity for innovation and scalability in supporting AI workloads. As AI applications continue to evolve, the selection of sophisticated, capable storage systems becomes a linchpin for the success and efficiency of AI-driven enterprises. This article brings together this entire series, sponsored by Solidigm.

3

LLMs are high-powered probabilistic next word generators, so the quality of the data they’re exposed to has a direct impact on the results that they generate. #datascience #AI #artificialintelligence https://hubs.li/Q02-WFfk0

1

🔍 Can a single platform change the way enterprises handle data and AI? 💡 Snowflake is making waves with Cortex AI, multimodal LLMs, and Hybrid Tables. 🚀 What’s behind the multimodal insights powering these new AI tools?  And how do they tackle unstructured data with seamless governance? 🔒 Plus, did someone say leaked password alerts for enterprise safety? 😱 🌐 Dive into how these innovations are rewriting the rules of secure, real-time analytics. ⏳ Are we entering the next era of decision-making with AI at its core? 👉 Tap the link to uncover how Snowflake is setting the new standard for enterprise AI. https://lnkd.in/deUcNMW8 GigaNectar #AI #DataManagement #Snowflake #EnterpriseTech #TechUpdates

1

1 Comment