Bob DuCharme

Included in this article is the mention of vector databases. Unsurprisingly, Michael Stonebraker and Andy Pavlo feel there is not a need for vector-specific databases. They believe relational databases will adopt their features, which is already happening. Check out the details here: https://lnkd.in/gyq_4WjP

27

1 Comment

Very insightful work by Anthropic’s interpretability team. And an amazing paper, with outstanding writing and figures. The idea is very simple: interpret LLMs by leveraging sparse autoencoders as surrogate models of the MLP of transformer blocks, which allow one to disambiguate the superposition of features captured by a single neuron. A simple idea, but a very careful and complex execution, as it is often the case in our line of work. The paper goes into many details and provide a large array of insights, although the gist of the implementation remains obfuscated due to the closed source nature of Claude. Too bad, because this is the kind of work that we need to better understand and eventually trust LLMs. This is demonstrated by the authors in the section ‘Influence on Behavior’, where they show that clamping some features to either high or low value during inference is “remarkably effective at modifying model outputs in specific, interpretable ways”. Hopefully this kind of work is going to be replicated and generalized to open-weights models, such that we have new ways to steer their behavior. https://lnkd.in/eVym7f_f #interpretability #xai #explainableai #steerableai #anthropic #claude #anthropic

3

this demo never gets old for me... check out claude 3.5 sonnet and its new multi-modal capability especially with interpreting charts, graphs, or transcribing text from imperfect images.

51

1 Comment

This week in #bigdata -- a #Postgres data lakehouse; top #dataannotation and #datalabeling platforms; and the big #GenAI request from America's Data Agency. And much, much more on www.datanami.com.

12

The 27Global Engineering blog goes deep into our technical solution design for clients. Check out our recent post by Sun Jang on the Databricks Unity Catalog and how we implemented it in multiple enviroments. https://lnkd.in/gBjFfySa

18

1 Comment

To foster the growth and maturity of AI Governance, YDC has open sourced its AI Governance artifacts including components, controls, regulations and citations on Hugging Face. Our intention is to collaborate with the market to extend these #aigovernance artifacts over time. These artifacts may be simply copied for internal use or accessed via the Hugging Face API. You can then enrich these artifacts in third party applications like Microsoft Purview to render an AI Governance Maturity Assessment. Read more here https://lnkd.in/erUKKtpt

34

5 Comments

Anthropic introduced their most intelligent and powerful model to date, Claude 3.5 Sonnet. It is now available in Amazon Bedrock. Claude 3.5 Sonnet outperforms Anthropic’s previous top-tier Claude 3 Opus model across a wide range of industry benchmarks, including graduate-level expert reasoning, code generation, and more. But the key aspect is its price-performance advantage. This new model delivers enhanced capabilities — and at only one-fifth the cost of Claude 3 Opus, empowering customers to create cutting-edge, generative AI-powered solutions at a fraction of the cost. https://lnkd.in/dgVN6EhZ Really interesting time...

21

OCLC recently announced a comprehensive strategy to bring linked data into mainstream library cataloging workflows. It’s a long-term approach, recognizing that most libraries will move to linked data slowly and incrementally. At OCLC we’re committed to providing tools and resources to support the transition for everyone. We are working closely with libraries around the world, and know that staff at some libraries are already educating themselves on the topic while others have a lot of questions. To help, OCLC is rolling out linked data infrastructure and services that meet libraries where they are today and provide meaningful improvement to challenges facing libraries. Read more about it in this OCLC Next blog: https://lnkd.in/d5DWUMyf

90

2 Comments

Tables in PDFs are very prevalent and for those of us who are intent on extracting information from PDFs, they often contain very useful information in a very dense format. Processing tables in PDFs, if not the top 10 priority for humanity, it certainly is an important one. In this blog post, yours truly talks about how you can go from PDF tables to clean JSON data in minutes with the power of LLMWhisperer, Unstract's raw text extraction service. Read here: https://lnkd.in/g-_GsDFf #llms #pdf #dataextraction #unstructureddata #ai

41

Cirata simplifies Gerrit instance data replication Across multiple global sites http://mvnt.us/m2417326 @CirataData #DataCapture #DevOps #Data

2

And here they are. IBM's open source code LLMs with Apache 2.0 licensing: https://lnkd.in/gpJkaYvF In 3b/8b/20b/34b sizes and with instruction-tuned versions too. Particularly exciting for us at IBM Research India given our deep involvement in development of these models all the way from curation of the data (and that continues to expand) to training, and to additional downstream alignment and infusion into some of the coolest code assistant products in the market - watsonx Code Assistant for Z and for Ansible Lightspeed. Catch up on all the technical detail here on the granite code LLM family: https://lnkd.in/g4cu6Z4Z #ibmgranite #ibmresearchindia

152

2 Comments

Graph databases have emerged as a powerful tool for managing interconnected data, but not all are created equal. Join industry expert George Anadiotis and George Demarest as they uncover the key criteria for evaluating #graphdatabases in this #webinar. https://lnkd.in/eWBYYY6S

4

A coworker of mine wrote a great blog about metadata. Metadata is one of those concepts it seems like everyone knows, but few actually do! This blog goes over some core concepts that are very useful: https://bit.ly/4beCg2C #metadata #informationsecurity #datamangement #OptimussbrBeBold

20

Vector Databases the cornerstone of RAG. When scoping data design solutions this is a key consideration for future approaches to LLMs and other modern tooling. #data #database #architecture #datadesign #llm #ai #rag #analytics https://lnkd.in/eyHD5wHv

1

New on Runtime: As the data nerds prepare to take over San Francisco for the next two weeks at back-to-back conferences held by Snowflake and Databricks, a look at how dueling open-source storage formats are upending the data business. Just a few years ago choosing a format was a consequential decision, but that's changing as companies build data analytics services that work across formats. And that could move the competition beyond format wars to features like security and governance. With thoughts from Databricks' Jonathan Keller, Snowflake's Christian Kleinerman, Starburst's Justin Borgman, Theory Ventures' Tomasz Tunguz, and Microsoft's Arun Ulag.

36

It has been a common refrain lately that says "𝙂𝙤𝙤𝙙 𝘼𝙄 𝙙𝙚𝙥𝙚𝙣𝙙𝙨 𝙤𝙣 𝙜𝙤𝙤𝙙 𝙄𝘼", but few have provided examples. This certainly will. I've been noodling on the potential convergence of content (document) modeling and knowledge modeling. Information Architects (generally) haven't considered KM modeling as they design content models, but the emergence of neuro-symbolic AI RAG to construct deterministic solutions might open things up to some degree of convergence. Here are some of my latest thoughts and musings on the topic.

62

10 Comments