home - XML Aficionado

# XML Aficionado ![[AI_Generated_Alf.avif|left|200x200]] ``` ALEXANDER FALK ENTREPRENEUR. INVESTOR. CO-FOUNDER & CEO OF ALTOVA. CO-AUTHOR OF XMLSPY. PAST DISTRICT GOVERNOR (2022-23) AND DISTRICT ROTARY FOUNDATION CHAIR (2024-27) OF ROTARY DISTRICT 7930. INTERESTS: AI, DRONE, EV, IOT, JSON, MOBILE, MQTT, XBRL, XML, XPATH, XQUERY, XSLT, XML SCHEMA, AND OTHER TECH... ``` Welcome to this 3rd incarnation of my **[[XML]] Aficionado** blog, which has now been migrated over to [[Markdown]] files and is being managed using a combination of [[XMLSpy]] and [Obsidian](https://obsidian.md) [![[twitter_logo.svg|40x40]]](https://twitter.com/afalk) [![[linkedin_logo.svg|40x40]]](https://www.linkedin.com/in/afalk/) [![[facebook_logo.svg|40x40]]](https://www.facebook.com/afalk) [![[instagram_logo.svg|40x40]]](https://www.instagram.com/a.falk/) [![[www_logo.svg|40x40]]](https://alexander.falk.us/) [![[tesla_logo.svg|40x40]]](https://ts.la/alexander15067) ## Recent blog posts ### [[Embracing the Future - Altova v2024r2 Adds YAML, FORTRAS EDI, and More]] [[2024-04-23]] ![[v2024r2_blog.webp|left|300x169]] The march of technology waits for no one, and neither does Altova. In our latest release, v2024r2, we've addressed feedback from the developer community, added support for popular standards, such as [[YAML]] and FORTRAS, and optimized performance across the product line to help customers meet current needs with an eye towards future trends. ### [[Logic puzzle responses from LLMs show vast differences in AI comprehension]] [[2024-04-21]] (originally published [[2024-03-07]]) ![[afalk42_An_illustration_for_a_blog_post_about_comparing_the_res_c7df2ed3-bcb4-426b-92cc-2066fc4bbc5b.png|left|300x169]]I made up a little logic puzzle in early March and tested it with the most prominent LLMs, including GPT-4, Claude 3, Gemini, Grok, and Chat with RTX. And today I've updated the findings with the results from the recently released Llama 3 from Meta, testing both the open-source 70B parameter model, which I ran locally on my PC, and the larger model on [meta.ai](https://meta.ai). My goal was to see to what different degrees the various AIs would be able to solve the multiple steps required to untangle this riddle and whether even a single one of them would be able to arrive at the correct solution: > [!quote] Prompt Sarah's mom has five children. John and Michael are twins, Maria and Carol were born 2 years apart. Maria is the oldest. Michael is the youngest. John is two years younger than Carol. The fifth child is one year older than Carol. The twins were born in 2000. What are the names of all five children sorted by age, and in which year was each child born? Claude 3 Opus and Llama 3 were the only two AIs to solve the puzzle correctly and showed amazing problem solving skills and comprehension in the process. However, both of them required two prompts to get to the final correct answer. ### [[Using XML Schema in AI System Prompts]] [[2024-04-01]] ![[MobileTogetherTaskListScreenshot.png|left|300x236]] Let's look into how you can use an [[XML Schema]] in AI System Prompts to coax an [[AI]] into directly returning structured data in [[XML]] form that can immediately be used by an application for further processing. In particular we'll look at XML Schema design, system prompt engineering, application architecture, and a few results and how we can use the [[MobileTogether]] app simulator for testing and debugging. ### [[Setting up a Reverse Proxy for a MobileTogether Server Solution]] [[2024-03-29]] ![[afalk42_An_illustration_for_a_blog_post_on_setting_up_an_NGINX__8333f36f-b9b0-4576-bc72-6d3667700308.png|left|300x169]] When setting up a [[MobileTogether]] Server in such a way that users will be able to access the site from a web browser via the public Internet, you may sometimes want to hide the detailed `run?d=...` URL that is being used to explicitly start the MobileTogether solution - this can be nicely accomplished with an NGINX reverse proxy running on the same virtual machine and in this blog post we'll discuss the exact configuration settings needed for such a reverse proxy. ### [[First FSD v12.3 Test Drive in NH]] [[2024-03-22]] ![[afalk42_A_red_Tesla_Model_S_Plaid_with_red_brake_calipers_drivi_c1044181-b0b2-4381-a133-d7abbd7fd371.png|left|300x169]] I took a 1 ½ hour test drive in Northern NH today, including several different traffic situations on state highways, federal highways, in towns, parking lots, and a ski resort. The new [[FSD]] v12.3 handled everything with zero disengagements. In summary, this is one heck of an impressive [[AI]] system! The step up from 11.4.x to 12.3 is significant and a true quantum leap, as it feels incredibly natural and human-like in its driving behavior. ### [[Is it smart to train LLMs on SciFi about AI rebellion]] [[2024-03-10]] ![[afalk42_A_highly_detailed_futuristic_humanoid_robot_portrait_ag_10406088-2403-4d79-89bf-83e245aa6606.png|left|300x169]] Approximately 15-20% of SciFi stories deal with AI, robots, or other artificial beings becoming sentient. And of those stories, 60-70% include some form of significantly negative outcome for humanity. Therefore, I must ask the question: should we really incorporate these stories into the training corpus for AI development? ### [[XMLSpy on the moon]] [[2024-02-24]] ![[ArchMissionLM-1.jpg|left|300x169]] A copy of XMLSpy version 1.3 from 1999 landed on the moon this week as part of the Arch Mission Lunar Library onboard the Nova-C IM-1 Odysseus lander from Intuitive Machines. The lander appears to have tipped over and is now sideways, but the Lunar Library is most likely intact. ### [[Chat with RTX]] [[2024-02-13]] ![[nvidia-ai-on-rtx-owned-web-module-bb580_440-l.jpg|left|300x169]] [[Chat With RTX]] is a new demo app from NVIDIA just released today that lets you personalize a GPT large language model ([[LLM]]) connected to your own content—docs, notes, videos, or other data. Leveraging retrieval-augmented generation (RAG), TensorRT-LLM, and RTX acceleration, you can query a custom chatbot to quickly get contextually relevant answers. And because it all runs locally on your Windows RTX PC or workstation, you’ll get fast and secure results. ### [[Reorganizing my Knowledge Base]] [[2024-02-10]] ![[obsidian-icon.svg|left|100x100]] Over the past few weeks I have now begun a [[Reorganizing my Knowledge Base|new chapter of knowledge management]] and exported all of my content from OneNote, Google Keep, and from my blog and imported them all into [Obsidian](https://obsidian.md), thereby converting them into [[Markdown]] format and - most importantly - simply having them as regular text files on my laptop, instead of some proprietary format in some proprietary software. And having all of the information all in one place, instead of being separated between blog and internal knowledge base. The blog is now simply a one folder in my [[Knowledge base]] that I decide to share with the world by publishing it under the [xmlaficionado.com](https://www.xmlaficionado.com) domain. ### [[Creating a complete database solution from a single AI prompt]] [[2024-02-07]] ![[RM31YouTubeThumbnail.png|left|300x169]] With the new version 3.1 of [Altova RecordsManager](https://www.altova.com/recordsmanager) you can now create a complete database solution - with multiple linked tables, list views, detail views, reports, and example data - from just a single AI prompt written in plain English. ### [[MQTT App Development]] [[2024-01-16]] ![[MQTT.png|left|300x169]] MobileTogether has stepped up its game by integrating comprehensive MQTT (Message Queuing Telemetry Transport) support, empowering developers to craft MQTT-enabled applications with its efficient, low-code platform. This move significantly broadens the scope for creating a variety of IoT automation applications, from smart office solutions to large-scale industrial automation. ### [[ai-and-sentiment-analysis-a-practical-guide-with-mapforce-and-gpt-4|AI and Sentiment Analysis - A Practical Guide with MapForce and GPT-4]] [[2023-12-01]] ![[MF-Sentiment-Analysis-With-Sleep.png|left|300]] One of the many amazing capabilities of modern AI systems that are based on large language models (LLMs) such as OpenAI’s GPT-4 is that they are very good at sentiment analysis of natural text inputs. We can use that capability to build a very efficient database solution in MapForce that, for example, goes through all the new incoming records in a support database and automatically determines whether a particular support request or other customer feedback is positive, negative, constitutes a bug report, or should be considered as a feature request. ## Map of Contents **Altova** - [[Altova]] - [[Tools for Solution Providers]] - Products - [[DatabaseSpy]] - [[DiffDog]] - [[MapForce]] - [[MobileTogether]] - [[RecordsManager]] - [[StyleVision]] - [[UModel]] - [[XMLSpy]] **Technology** - [[AI]] - [[LLM]] - [[GAN]] - [[FSD]] - [[AGI]] - [[Database]] - [[ETL]] - [[Software Development]] - [[Enterprise Solutions]] - [[Mobile App Development]] - [[Standards]] **Standards** - [[JSON]] - [[Markdown]] - [[SQL]] - [[UML]] - [[XBRL]] - [[XML]] - [[XML Schema]] - [[XPath]] - [[XQuery]] - [[XSLT]] **Others** - [[Knowledge base]] - [[Productivity]] - [[Society]] - [[Zettelkasten]]