This week we feature articles from Sayash Kapoor & Arvind Narayanan, and Daniil Gurgurov, Tanja Bäumel & Tatiana Anikina.

Additional reading comes from Sanjeev Mohan, Amber Case, and Michael Andrews.

News comes from RWS, GraphAware & Senzing, Elastic, and Box.

Our next issue will arrive July 17.

All previous issues are available at https://gilbane.com/gilbane-advisor-index

---

Opinion / Analysis

AI scaling myths

Scaling will run out. The question is when.

Sayash Kapoor and Arvind Narayanan shed light on a number of misconceptions or unstudied assumptions while answering the question.  (8 min)

https://www.aisnakeoil.com/p/ai-scaling-myths

Multilingual large language models and curse of multilinguality

“This paper aims to provide a brief overview of the architectures of the most prominent multilingual LLMs, including details such as their pre-training objective functions, data sources, tokenization schemas, the number of languages supported, and the peculiarities of each individual multilingual LLM. Subsequently, the primary challenge facing multilingual LLMs, known as the “curse of multilinguality” Conneau et al. (2020), and the current attempts to solve it, are discussed.”

This paper by Daniil Gurgurov, Tanja Bäumel, and Tatiana Anikina is a comprehensive and current update on multilingual LLM tools and challenges. (18 min)

Semantic Scholar link: https://www.semanticscholar.org/reader/2235420149c8f839e29d4cf8df1894d7a9f3de37
arXiv HTML and clean PDF link: https://arxiv.org/html/2406.10602v1

More Reading

• Coping with copies: Strategic dimensions of reuse and duplication via Story Needle
• Useful analysis with references to Snowflake… Databricks Data & AI 2024 Summit via Sanjeev Mohan
• Reducing cognitive load… Apple’s approach to AI should rewrite the industry’s expectations for the technology via Amber Case

All Gilbane Advisor issues

---

Content technology news

RWS launches Trados Studio 2024

The latest version includes advanced AI capabilities and a plethora of new features to boost productivity and increase translation quality.
https://www.trados.com

Elastic introduces Playground to accelerate RAG development with Elasticsearch

The new interface enables developers to iterate and build RAG applications for A/B testing LLMs, tuning prompts and chunking data.
https://www.elastic.co/search-labs/blog/rag-playground-introduction

Graph analytics enhanced by GraphAware & Senzing

The partnership integrates Senzing entity resolution capabilities into GraphAware Hume, GraphAware’s graph-based enterprise intelligence platform.
https://senzing.com/graph-analytics-graphaware/

Box enhances Box AI for intelligent content management

Enterprises to get unlimited access to Box AI in Notes, Documents, and Hubs, GPT-4o Integration in Hubs, new file types, and developer tools.
https://blog.box.com/box-announces-powerful-enhancements-box-ai-intelligent-content-management

All content technology news

---

The Gilbane Advisor is authored by Frank Gilbane and is ad-free, cost-free, and curated for content, computing, web, data, and digital experience technology and information professionals. We publish recommended articles and content technology news most Wednesdays. We do not sell or share personal data.

Subscribe | View online | Editorial policy | Privacy policy | Contact

This week we feature articles from Eugene Yan, Bryan Bischof, Charles Frye, Hamel Husain, Jason Liu & Shreya Shankar, and Patrick Walsh.

Additional reading comes from Mel Richey, Emma Roth, and Scott Brinker.

News comes from Adobe, Argo Translation, Netlify, Acrolinx.

Next issue arrives July 3.

All previous issues are available at https://gilbane.com/gilbane-advisor-index

---

Opinion / Analysis

What we learned from a year of building with LLMs (Part I – tactical)

“This is the first of three pieces. It dives into the tactical nuts and bolts of working with LLMs. We share best practices and common pitfalls around prompting, setting up retrieval-augmented generation, applying flow engineering, and evaluation and monitoring.”

Eugene Yan, Bryan Bischof, Charles Frye, Hamel Husain, Jason Liu and Shreya Shankar are experienced practitioners in a variety of roles and projects that collaborated to create a unique and valuable lessons-learned resource. Links to Parts 2 – operational, and Part 3 – strategic, are included. (30 min)

https://www.oreilly.com/radar/what-we-learned-from-a-year-of-building-with-llms-part-i

Analysis of Apple’s new AI Private Compute Cloud

As others have noted, there has been a lot of confusion about Apple’s Private Compute Cloud (PCC) architecture. Patrick Walsh provides an easily readable and balanced analysis – well worth a read. (12 min)

https://blog.ironcorelabs.com/analysis-of-apples-new-ai-private-compute-cloud-14f3e16105c9

More Reading

• True fact! APIs are “very important” to martech buyers today, but they’ll be crucial for the coming wave of AI agents via chiefmartec
• Good description of why it’s hard but worth it… Entity-Resolved knowledge graphs & data fusion via Towards Data Science
• US sues Adobe for ‘deceiving’ subscriptions that are too hard to cancel. They’re not the only ones. Someone must have a list? via The Verge

All Gilbane Advisor issues

---

Content technology news

Adobe updates Acrobat with GenAI and chat across document types

Adobe brings Firefly features into Acrobat so customers can generate and edit images directly in PDFs, and multi-document chat to AI Assistant.
https://news.adobe.com/news/news-details/2024/Adobe-Reimagines-Acrobat-Bringing-Firefly-AI-to-PDFs-and-Expanding-Use-Across-More-Document-Types/default.aspx

Argo Translation launches CMSConnect: Slack

AI technology translates Slack messages in real-time, improving global team and customer support productivity.
https://www.argotrans.com/technology/cmsconnect-slack

Netlify announces Adobe Experience Manager headless integration

AEM users can leverage the Netlify headless CMS to streamline workflows and reduce the touchpoints needed across multi-channel projects.
https://www.netlify.com/press/netlify-announces-adobe-experience-manager-headless-integration/

Acrolinx launches new AI capabilities

Get Suggestions, AI Assistant, and AI Guardrails make content creation easier, faster, and compliant with your enterprise content standards.
https://www.acrolinx.com

All content technology news

---

The Gilbane Advisor is authored by Frank Gilbane and is ad-free, cost-free, and curated for content, computing, web, data, and digital experience technology and information professionals. We publish recommended articles and content technology news most Wednesdays. We do not sell or share personal data.

Subscribe | View online | Editorial policy | Privacy policy | Contact

This week we feature articles from Louise de Leyritz, and Ben Thompson.

Additional reading comes from Mike Loukides, Michael Andrews, and Vik Paruchuri, Samuel Lampa & Trey Saddler.

News comes from Ontotext, Contentstack, Databricks, and Snowflake.

Our next issue will arrive June 26.

All previous issues are available at https://gilbane.com/gilbane-advisor-index

---

Opinion / Analysis

AI assistants, copilots, and agents in data & analytics: What’s the difference?

Understanding AI autonomy: assistants, copilots, agents, and their impact on business value

Louise de Leyritz provides a well-written piece especially valuable for non-technical management, but also useful for many others across your organization. (7 min)

https://towardsdatascience.com/ai-assistants-copilots-and-agents-in-data-analytics-whats-the-difference-2e63f8fb2384

Apple Intelligence is right on time

Ben Thompson published this article just before Apple’s announcements a couple of days ago after an apparent, and accurate, leak. Thompson’s analysis remains on target and is a good read. (16 min)

https://stratechery.com/2024/apple-intelligence-is-right-on-time

More Reading

• What O’Reilly is building… Answers: generative AI as learning tool via O’Reilly Radar
• Marker converts PDFs to Markdown using deep learning via Vik Paruchuri
• Tracking content’s history: the key to better content maintenance via Story Needle

All Gilbane Advisor issues

---

Content technology news

Ontotext announces Metadata Studio 3.8

Metadata Studio (OMDS) is a tool designed for knowledge graph enrichment through text analytics of unstructured documents.
https://www.ontotext.com/products/ontotext-metadata-studio/

Contentstack updates personalization tools

Extends headless CMS with new capabilities to overcome the personalization challenges of delivering highly targeted, branded content.
https://www.contentstack.com/solutions/personalization

Databricks to acquire Tabular

With support for Apache Iceberg and Linux Foundation Delta Lake lakehouse formats organizations are not limited by formats their data is in.
https://www.databricks.com/company/newsroom/press-releases/databricks-agrees-acquire-tabular-company-founded-original-creators

Snowflake announces enhancements to Snowflake Cortex AI,  Snowflake ML, and more

Additional announcements include Snowflake Copilot, Cortex Guard, Document AI, Hybrid & Iceberg Tables, Guard, Trail, and Pandas API.
https://www.snowflake.com/news/snowflake-brings-industry-leading-enterprise-ai-to-even-more-users-with-new-advancements-to-snowflake-cortex-ai-and-snowflake-ml/

All content technology news

---

The Gilbane Advisor is authored by Frank Gilbane and is ad-free, cost-free, and curated for content, computing, web, data, and digital experience technology and information professionals. We publish recommended articles and content technology news most Wednesdays. We do not sell or share personal data.

Subscribe | View online | Editorial policy | Privacy policy | Contact

This week we feature articles from Steve Jones, and Arvind Narayanan & Sayash Kapoor.

Additional reading comes from Victor Morgante, Andrej Karpathy, Daniel Tunkelang, and LlamaIndex.

News comes from Sinequa, Perplexity, Tonic.ai, and Siteimprove.

Our next issue will arrive May 12.

All previous issues are available at https://gilbane.com/gilbane-advisor-index

---

Opinion / Analysis

The AGI hype is damaging the AI opportunity

“There are two ways that this hype is damaging the view of AI within businesses today.

• It sounds like utter snake oil promising that in future it will be great
• It implies that current AI approaches aren’t good enough”

Steve Jones calls out the hypesters. (6 min)

https://blog.metamirror.io/the-agi-hype-is-damaging-the-ai-opportunity-c45bae1f7f5a

Scientists should use AI as a tool, not an oracle

AI hype damage is not just bad for business, it’s bad for science. Arvind Narayanan & Sayash Kapoor explain “How AI hype leads to flawed research that fuels more hype”. (6 min)

https://www.aisnakeoil.com/p/scientists-should-use-ai-as-a-tool

More Reading

• Multi-model, GQL… The future of databases — graph relational via Towards Data Science
• Property Graph Index: A new way to build knowledge graphs with LLMs via LlamaIndex Blog 
• A real deal… Reproducing GPT-2 (124M) in llm.c in 90 minutes for $20 via Andrej Karpathy
• Bags of documents or, bags of queries as sparse document representations via Daniel Tunkelang

All Gilbane Advisor issues

---

Content technology news

Sinequa releases new generative AI assistants

The enterprise domain-specific generative AI assistants integrate with content and applications to augment and transform knowledge work.
https://www.sinequa.com/company/press/sinequa-augments-companies-with-release-of-new-generative-ai-assistants/

Perplexity introduces Perplexity Pages

Pages lets you create, organize, and share information. Search any topic, and instantly receive a well-structured, beautifully formatted article.
https://www.perplexity.ai/page/new

Tonic.ai launches secure unstructured data lakehouse for LLMs

Tonic Textual is specifically architected to meet the complexity, scale, integration and privacy demands of enterprise unstructured data.
https://www.tonic.ai/textual

Siteimprove launches new product features

New capabilities designed to turn data into easy to understand insights, increase cross-organizational collaboration & facilitate decision making.
https://www.siteimprove.com/hello/new-and-siteimproved-2024-q2-product-release/

All content technology news

---

The Gilbane Advisor is authored by Frank Gilbane and is ad-free, cost-free, and curated for content, computing, web, data, and digital experience technology and information professionals. We publish recommended articles and content technology news most Wednesdays. We do not sell or share personal data.

Subscribe | View online | Editorial policy | Privacy policy | Contact

This week we feature articles from Sunila Gollapudi, and Leontien Talboom & Chris Knowles.

Additional reading comes from Heather Hedden, Cassie Kozyrkov, and Jim Clyde Monge.

News comes from Elastic, DataStax, Flatfile, and Foxit & Straker Translations.

Note: We’ll be off next week, back on June 5th.

All previous issues are available at https://gilbane.com/gilbane-advisor-index

---

Opinion / Analysis

Combine text embeddings and knowledge (graph) embeddings in RAG systems

Sunila Gollapudi provides a good introduction and how-to suitable for technical and not so technical readers.

“In this article, I am excited to present my experiments combining Text Embeddings and Knowledge (Graph) Embeddings and observations on RAG performance. I will start by explaining the concept of Text and Knowledge Embeddings independently, using simple open frameworks, then, we will see how to use both in RAG applications.” (15 min)

https://towardsdatascience.com/combine-text-embeddings-and-knowledge-graph-embeddings-in-rag-systems-5e6d7e493925

Raw flux streams and obscure formats: Further work around imaging 5.25-inch floppy disks

I’m sure the subject has some of you shaking your heads for any number of reasons. But for those connected with digital preservation efforts, this case-study/lessons-learned piece from Leontien Talboom & Chris Knowles at Cambridge University could be very helpful. Some of the comments may also be useful. The just-curious may be shocked at the complexity involved. (8 min)

https://digitalpreservation-blog.lib.cam.ac.uk/raw-flux-streams-and-obscure-formats-further-work-around-imaging-5-25-inch-floppy-disks-5a2cf2e5f0d1

More Reading

• All the announcements at Microsoft Build – Book of News via Microsoft
• What are Synonym Rings (or Search Thesaurus)? via The Accidental Taxonomist
• Why we urgently need AI control systems via Cassie Kozyrkov
• Useful quick summary… 5 major announcements in Google IO 2024 via Generative AI
• Apple’s upcoming accessibility features, including eye tracking, music haptics, and vocal shortcuts via Apple

All Gilbane Advisor issues

---

Content technology news

DataStax launches new Hyper-Converged Data Platform

Brings OpenSearch and Apache Pulsar to HCD Platform; DataStax Enterprise 6.9 enables self-managed data workloads for GenAI.
https://www.datastax.com/press-release/datastax-launches-new-hyper-converged-data-platform-giving-enterprises-the-complete-modern-data-center-suite-ceeded-for-ai-in-production

Elastic announced Search AI Lake to scale low latency search

Architecture optimized for real-time, low-latency applications including search, retrieval augmented generation (RAG), observability & security.
https://ir.elastic.co/news/news-details/2024/Elastic-Announces-First-of-its-kind-Search-AI-Lake-to-Scale-Low-Latency-Search/default.aspx

Flatfile unveils new AI-powered data transformation features

Data transformation and data migration capabilities for business users, data analysts, systems integration teams, and enterprise developers.
https://flatfile.com/news/flatfile-unveils-ai-powered-data-transformation/

Foxit partners with Straker Translations

The collaboration adds translation capabilities to Foxit’s eSignature services, enabling users to translate and sign documents in multiple languages.
https://www.foxit.com ■ https://www.straker.ai

All content technology news

---

The Gilbane Advisor is authored by Frank Gilbane and is ad-free, cost-free, and curated for content, computing, web, data, and digital experience technology and information professionals. We publish recommended articles and content technology news most Wednesdays. We do not sell or share personal data.

Subscribe | View online | Editorial policy | Privacy policy | Contact

This week we feature articles from Alex Klein, and Alberto Romero.

Additional reading comes from Jim Clyde Monge, James O’Donnell, Cobus Greyling, and Scott Brinker.

News comes from Adobe, Apple, SoundHound & Perplexity, and Acquia.

All previous issues are available at https://gilbane.com/gilbane-advisor-index

---

Opinion / Analysis

The agentic era of UX

“The future of digital experience is here — but it’s being minced into microscopic use cases.”

Alex Klein argues that AI user experience requires a holistic approach that pulls together point use cases into a workflow that addresses more complete user journeys, and that the role of design is critical. He explains, and has some good advice on how to proceed. (8 min)

https://uxdesign.cc/the-agentic-era-of-ux-4b58634e410b

OpenAI rules the changes but Meta changes the rules

Alberto Romero comes up with an enlightening analysis of the AI market focused on Meta’s strategy of making LLMs a commodity with Llama, and OpenAi, representing current LLM leadership. 

OpenAI’s Monday announcement of GPT-4o keeps the pressure on Google, Anthropic, et al. Some the impressive GPT-4o demos were on phones and suggest what we might see if Apple and OpenAI finalize an agreement. (7 min)

https://www.thealgorithmicbridge.com/p/openai-rules-the-changes-but-meta

More Reading

• Questions… Stack Overflow partners with OpenAI via Generative AI
• Different data… The robot race is fueling a fight for training data via MIT Technology Review
• Stack update… Large language model (LLM) stack — version 6 via Cobus Greyling
• Pretty stacks… 26 magnificently illustrated martech stacks from the 2024 Stackie Awards via chiefmartec

All Gilbane Advisor issues

---

Content technology news

Adobe launches Acrobat AI Assistant for the enterprise

Acrobat AI Assistant can be deployed in minutes with enterprise controls, increasing productivity for knowledge workers on desktop, web and mobile.
https://blog.adobe.com/en/publish/2024/05/08/adobe-acrobat-ai-assistant-enterprise-our-commitment-data-governance-security

SoundHound AI and Perplexity partner

Perplexity’s capabilities added to SoundHound Chat AI will respond to questions conversationally with real-time knowledge from the web.
https://www.soundhound.com/newsroom/press-releases/soundhound-ai-and-perplexity-partner-to-bring-online-llms-to-its-next-gen-voice-assistants-across-cars-and-iot-devices/■ https://www.perplexity.ai

Apple Final Cut Pro 2 on iPad and Mac get new AI features

Final Cut Pro 2 transforms video creation with Live Multicam on the new iPad Pro and new AI features on iPad Pro and Final Cut Pro Mac 10.8.
https://www.apple.com/newsroom/2024/05/final-cut-pro-transforms-video-creation-with-live-multicam-on-ipad-and-new-ai-features-on-mac/

Acquia supports latest W3C content accessibility guidelines

With the Accessibility Module in Monsido, customers can scan their websites, obtain an accessibility score, and get recommendations to improve accessibility.
https://www.acquia.com/products/monsido

All content technology news

---

The Gilbane Advisor is authored by Frank Gilbane and is ad-free, cost-free, and curated for content, computing, web, data, and digital experience technology and information professionals. We publish recommended articles and content technology news weekly. We do not sell or share personal data.

Subscribe | View online | Editorial policy | Privacy policy | Contact

This week we feature articles from Rachel Draelos, and Benedict Evans.

Additional reading comes from Walid Saba, Benj Edwards, and Daniel Tunkelang.

News comes from WordPress VIP, Trados, Snowflake, and AtScale.

Note: Taking next week off — next issue on May 15th.

All previous issues are available at https://gilbane.com/gilbane-advisor-index

---

Opinion / Analysis

Human and artificial general intelligence arises from next token prediction

“This post frames and explores the hypothesis that general intelligence arises when a learning system becomes very good at next token prediction. This hypothesis is often implied, hidden, or dancing at the margins of industrial and academic AI research — but so far, I don’t think it’s received as much open discussion as I think it merits.”

Rachel Draelos, who is an MD with a PhD in Computer Science starts by pointing out that humans are next-word predictors — literally all the time. She then makes the more general case that we are token predicting the world around us through our senses, and how all of this is consistent with current neuroscience research. 

An interesting read and in spite of the topic the post is easily accessible to a general audience — there are links to technical papers. (16 min)

https://towardsdatascience.com/human-and-artificial-general-intelligence-arises-from-next-token-prediction-712500310a93

Looking for AI use-cases

“We’ve had ChatGPT for 18 months, but what’s it for? What are the use-cases? Why isn’t it useful for everyone, right now? Do Large Language Models become universal tools that can do ‘any’ task, or do we wrap them in single-purpose apps, and build thousands of new companies around that?”

Benedict Evans is trying to figure out how the AI application market will evolve beyond niche use cases to deliver on the widely expected growth and step change. His focus on product market fit makes his article more useful than speculative. (9 min)

https://www.ben-evans.com/benedictevans/2024/4/19/looking-for-ai-use-cases

More Reading

• June announcements could be interesting… Apple releases eight small AI language models aimed at on-device use via ars technica
• LLMs and RAG are great. What’s next? Because they’re not enough. via Daniel Tunkelang
• Walid Saba wants to know what you think… A refutation of John Searle’s famous Chinese Room Argument? via Ontologik

All Gilbane Advisor issues

---

Content technology news

WordPress VIP introduces VIP API Mesh to enable composable digital experiences

API Mesh simplifies data integration with a single API, improves caching, supports data transformation, and is accessible to technical and non-technical users.
https://wpvip.com/2024/04/25/vip-api-mesh-composable-digital-experiences/

Trados announces Trados Studio 2024

The computer-assisted translation tool improves accessibility, democratizes access to generative AI, and increases management capabilities.
https://slator.com/trados-enhances-platform-with-new-accessibility-ai-and-management-features/

Snowflake launches Arctic, an open enterprise-grade LLM

Snowflake Arctic LLM is a part of the Snowflake Arctic model family that also include practical text-embedding models for retrieval use cases.
https://www.snowflake.com/news/snowflake-launches-arctic-the-most-open-enterprise-grade-large-language-model/

AtScale introduces Developer Community Edition for semantic modeling

The Developer Community Edition is a fully featured, free version of AtScale’s semantic layer platform, promoting collaboration and versatility.
https://www.atscale.com/press/atscale-developer-community-edition-for-semantic-modeling/

All content technology news

---

The Gilbane Advisor is authored by Frank Gilbane and is ad-free, cost-free, and curated for content, computing, web, data, and digital experience technology and information professionals. We publish recommended articles and content technology news weekly. We do not sell or share personal data.

Subscribe | View online | Editorial policy | Privacy policy | Contact

This week we feature articles from Vincent Koc, and Mike Dillinger.

Additional reading comes from Stanford HAI, Kunal Sawarkar, Avi Chawla, and The Linux Foundation.

News comes from Adobe, NebulaGraph, ThoughtSpot, and expert ai.

All previous issues are available at https://gilbane.com/gilbane-advisor-index

---

Opinion / Analysis

The GenAI Compass: a UX framework to design generative AI experiences

Vincent Koc’s comprehensive framework is a must read for teams building GenAI applications. (17 min)

https://uxdesign.cc/the-genai-compass-a-ux-framework-to-design-generative-ai-experiences-49a7d797c114

Continuous Knowledge Graphs for Neurosymbolic AI

“One blocker to more widespread deployment of knowledge graphs in AI is the belief that they are inherently a discrete form of knowledge. This is how they are often contrasted with large language models, which are described as a continuous form of knowledge. And it is a counterproductive way of characterizing these forms of knowledge representation that fundamentally misrepresents both.”

Mike Dillinger’s thoughtful and stimulating piece should be part of any discussion on the relative roles of LLMs and KGs. (8 min)

https://medium.com/@mike.dillinger/continuous-knowledge-graphs-for-neurosymbolic-ai-9000b4aa0eb7

More Reading

• The annual AI Index Report 2024 – Artificial Intelligence Index via Stanford HAI
• Short and useful… Llama3 is here – key takeaways via Towards Generative AI
• Continuous challenge… Linux Foundation launches Margo to deliver long-awaited edge interoperability via The Linux Foundation
• OpenAI or DIY? unveiling the true cost of self-hosting LLMs via VentureBeat

All Gilbane Advisor issues

---

Content technology news

Adobe adds Firefly to Adobe Express App

Adobe Express brings popular Firefly GenAI features like Text to Image, Generative Fill, Text Effects and Text to Template to mobile devices.
https://news.adobe.com/news/news-details/2024/All-New-Adobe-Express-Mobile-App-with-Firefly-AI-Now-Available-to-Millions-Empowering-them-to-Create-Standout-Content-On-the-Go/default.aspx

NebulaGraph Enterprise v5.0 offers native GQL support

Built on and designed for GQL data compatibility and interoperability to provide the business value of graph databases across various scenarios.
https://www.nebula-graph.io/posts/nebulagraph_enterprise_5.0_gql_support ■ https://www.iso.org/standard/76120.html

ThoughtSpot renames and adds features to ThoughtSpot Everywhere

Initiatives support developers and product builders to help their customers, partners, and employees with GenAI and embedded natural language search. 
https://www.thoughtspot.com/press-releases/thoughtspot-makes-embedding-ai-powered-analytics-easy-and-ubiquitous-for-everyone

Expert.ai launches Insight Engine for Life Sciences

Harnessing advanced AI capabilities to mine and aggregate scientific content, synthesize knowledge, extract relevant information & reveal hidden correlations.
https://www.expert.ai/expert-ai-launches-insight-engine-for-life-sciences/

All content technology news

---

The Gilbane Advisor is authored by Frank Gilbane and is ad-free, cost-free, and curated for content, computing, web, data, and digital experience technology and information professionals. We publish recommended articles and content technology news weekly. We do not sell or share personal data.

Subscribe | View online | Editorial policy | Privacy policy | Contact