{"id":140,"date":"2026-09-29T01:25:13","date_gmt":"2026-09-29T01:25:13","guid":{"rendered":"https:\/\/llama.gs\/blog\/?p=140"},"modified":"2026-09-29T01:25:13","modified_gmt":"2026-09-29T01:25:13","slug":"you-cannot-diagnose-what-you-dont-know","status":"publish","type":"post","link":"https:\/\/llama.gs\/blog\/index.php\/2026\/09\/29\/you-cannot-diagnose-what-you-dont-know\/","title":{"rendered":"You Cannot Diagnose What You Don’t Know"},"content":{"rendered":"\n

Even second graders know that AI generates slop by the gallon and is not to be trusted. They don’t know, though, that AI data centers will also produce enough e-waste to fill a string of shipping containers long enough to encircle the Earth six times at the equator<\/a>. At the same time when something as non-news worthy as a network router fails, AI goes down and a true AI-pocalypse occurs<\/a>. But, oh wait, what happened to AI taking over the world and trying to stop and kill everybody who would try to shut it down (so give me all your money so I can stop it and also build a bunker just in case)?<\/p>\n\n\n\n

Now, now, the AI believers will say. AI, besides consciousness will develop superb diagnostic and prognostic abilities and will predict all failures, human or machine. Here, sincerely yours, clears his throats sips his water and prepares his editor for writing a bit more about the process of diagnosis as a whole. Today we will talk about less mind boggling concepts such as AI consciousness and world domination and more about what it actually takes to perform real diagnosis.<\/p>\n\n\n\n

Let’s say that your boss comes to you and tasks you to diagnose something important: like a spaceship or motor rickshaw (doesn’t matter what, some system). Being cool and desiring to be gainfully employed for months and possibly until retirement you start talking about the multiple phases that will eventually result in something that is going to watch the spaceship or the motor rickshaw for failures (such as the dreaded engine maintenance light in your car). The phases are depicted in the diagram shown next.<\/p>\n\n\n\n

\"\"<\/figure>\n\n\n\n

We always start with some engineering description of what we are going to model. So we describe each component in the spacecraft (thrusters, antennas, tractor beam generators, etc.) in some formal or less formal language (think of Matlab, Python, or database schemas).<\/p>\n\n\n\n

What comes next and where LLMs and AI totally fail is fault-modes. So nobody designs devices (spaceships or not) to fail. Moreover, things hopefully fail relative rarely. And when they fail they typically do quickly and with a bang. And during this failure, it is probably your least priority to measure temperatures, pressure, and voltages and more like to duck and avoid the shrapnel.<\/p>\n\n\n\n

To run diagnosis we need to know how components fail. Since I proudly dissociate myself from the Californian AI companies, I will insist that all experiments are precise and reproducible. In other words and unlike the charlatan CEOs, I didn’t miss my science and engineering classes at college. The data collected is for building physics-inspired models or train machine-learning data, if you wish, but the most important aspect of this reproducibility is that we will be able to compare diagnostic approaches and systems in terms of Key Performance Indicators (KPIs), or diagnostic metrics if you wish. Everybody has heard about diagnostic metrics of various medical tests such as false-positives and false-negatives, but there are many more such as diagnostic accuracy, or even if you wish the amount of the super-expensive RAM due to the AI craze we have.<\/p>\n\n\n\n

Ex-Parrot evolved from work that I did for PARC ten years ago. PARC wanted to diagnose systems and I was interested in air conditioners (as I live in California, it is hot, and air conditioners fail a lot) but nobody would let me diagnose a real air conditioner. So I went to Walmart and bought the closest to an air-conditioner which turned out to be a hundred dollar refrigerator (I charged the company credit card). Then I took a pair of pliers, cut through many wires and instrumented the refrigerator. Of course, I wanted to know precise temperatures and voltages and other data, so I constructed a lot of hardware.<\/p>\n\n\n\n

\"\"<\/figure>\n\n\n\n

One of the main novelties of Ex-Parrot<\/a> is that it knows and one can schedule a diagnostic experiment. You just go to the web front-end, click a few buttons and start scheduling commands such as “toggle a relay” or “start a motor”. If the need arises, I will added the functionality to have LUA<\/a> scripts, right now it is just a linear sequence of commands.<\/p>\n\n\n\n

An Ex-Parrot diagnostic experiment is a sequence of commands. In the refrigerator use-case, for example, the scenario would do nothing let\u2019s say for four hours, then it would open the door to simulate a leak of thermal energy and then it would close it and again do nothing for some number of hours. The best part is that the whole experimental design can be done easily and with a tool that is shown next.<\/p>\n\n\n\n

\"\"
Neither obscure scripting skills nor agents are required to create a diagnostic experiment<\/figcaption><\/figure>\n\n\n\n

Running the script would execute the commands and produce dramatic videos like the one below.<\/p>\n\n\n\n

\"\"<\/figure>\n\n\n\n

The data is still available and can be used as a benchmark. I collected temperatures at many locations and even high-frequency voltage and current data. Soon there will be a public demo version of Ex-Parrot where users will be able to create a read-only account, browse the fridge data and even export it as HDF5 files for their own diagnostic and prognostic algorithms.<\/p>\n\n\n\n

Ex-Parrot<\/a> is to be combined with another software platform, called Psycho-Killer which takes the instrumented data and analyzes the performance of various diagnostic approaches.<\/p>\n\n\n\n

What will come next is a live demo of Ex-Parrot (read only, no actuation for demo accounts) and I am also developing a real sensing network with temperature and vibration sensors that will talk to Ex-Parrot via Matter. These sensors and the Ex-Parrot cloud service will be for sale so you can instrument your own devices.<\/p>\n\n\n\n

Before we close with the usual anti-AI chant, here is a taste of marketing and a bit of my plans. The old adage garbage-in, garbage-out is valid. Do you think you have a system that can benefit from good diagnostics and prognostics? Do you trust me that failure data is valuable and missing and if we capture it, we will be able to use it and sell it to other people who want to do diagnostics and prognostics? If so, give Llama Logic a call or drop a note and let’s do some quality diagnostic and prognostic engineering together!<\/p>\n\n\n\n

Ceterum censeo slopem esse delendam.<\/em><\/p>\n\n\n\n

(Cato the Elder ended every speech in the Roman Senate with \u201cCarthage must be destroyed\u201d \u2014 regardless of the topic. This is that, but for AI slop.)<\/p>\n\n\n\n

Repository:<\/strong> ex-parrot<\/a><\/p>\n\n\n\n

<\/p>\n","protected":false},"excerpt":{"rendered":"

Even second graders know that AI generates slop by the gallon and is not to be trusted. They don’t know, though, that AI data centers will also produce enough e-waste to fill a string of shipping containers long enough to encircle the Earth six times at the equator. At the same time when something as […]<\/p>\n","protected":false},"author":1,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[1],"tags":[],"class_list":["post-140","post","type-post","status-publish","format-standard","hentry","category-uncategorized"],"_links":{"self":[{"href":"https:\/\/llama.gs\/blog\/index.php\/wp-json\/wp\/v2\/posts\/140","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/llama.gs\/blog\/index.php\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/llama.gs\/blog\/index.php\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/llama.gs\/blog\/index.php\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/llama.gs\/blog\/index.php\/wp-json\/wp\/v2\/comments?post=140"}],"version-history":[{"count":12,"href":"https:\/\/llama.gs\/blog\/index.php\/wp-json\/wp\/v2\/posts\/140\/revisions"}],"predecessor-version":[{"id":158,"href":"https:\/\/llama.gs\/blog\/index.php\/wp-json\/wp\/v2\/posts\/140\/revisions\/158"}],"wp:attachment":[{"href":"https:\/\/llama.gs\/blog\/index.php\/wp-json\/wp\/v2\/media?parent=140"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/llama.gs\/blog\/index.php\/wp-json\/wp\/v2\/categories?post=140"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/llama.gs\/blog\/index.php\/wp-json\/wp\/v2\/tags?post=140"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}