A Geek Goes Flying

Walled garden as an illustrative concept We have a practice management system we rely on to manage our cases. It’s pretty good. It could be a lot better. There’s an API, which should enable integration with other tools that can help it be amazing. But. They locked it down. Annoying. They’re pumping money into their own home-grown AI tools. They’re trying to upsell me on their solutions. They boast they’ve put / are putting “tens of millions of dollars” into their home-grown AI tech. Cool. But, uh: “OpenAI and Anthropic expect to spend nearly $65 billion combined this year just on the costs to train and operate their AI models.”¹ Our vendor was showing off their agentic offering, and how it could calculate litigation deadlines based on, e.g., C.D. Cal. local rules . “Great, what about a judge's individual procedures?” E.g., Judge Walter  issues a standing order with a requirement not found in the local rules (which, annoyingly, is only in his Standing Order filed early in the c...

The jaunty sing-songy voice and music, the overly cute animation, the crass insertion into your nightmarescape of this dog water fountain ad ... I can’t even. (There’s one I just saw for GetQuietLab that’s in the same vein.) This is not the AI future we were promised. Or maybe it is. I don’t know anymore. 

An amp and a microphone One of the most frustrating parts about being a self-taught n00b when it comes to amp modeling (whether it be in a $100 NUX box or a bit of Fractal kit) is trying to decipher the different fake names the modelers have for the different pieces of kit, especially if you’re not already steeped in guitar tech. Like, the Jazz Clean NUX amp is (I understand) modeled after the Roland JC-120 Jazz Chorus . The Class A30, the  VOX AC30 , “a legendary British tube amp known for its jangly cleans, warm overdrive, and chimey tone that’s been a staple in rock, pop, and indie music.” Etc. Sometimes, there’s no hint. Using Fractal’s DynaCabs , for instance, you get a choice of four microphones for the Cab Impulse Response (IR) block. They’re helpfully</sarcasm> identified as Dynamic 1, Dynamic 2,  Condenser, and Ribbon.  If you read the Wiki , those map to the Shure SM57 and SM7B , the Soyuz 023 , and the Royer 121 – but that knowledge seems to be tribal ...

Per Slashdot : Raspberry Pi is launching a new add-on board capable of running generative AI models locally on the Raspberry Pi 5. Announced on Thursday, the $130 AI HAT+ 2 is an upgraded -- and more expensive -- version of the module launched last year, now offering 8GB of RAM and a Hailo 10H chip with 40 TOPS of AI performance. Once connected, the Raspberry Pi 5 will use the AI HAT+ 2 to handle AI-related workloads while leaving the main board's Arm CPU available to complete other tasks. Unlike the previous AI HAT+, which is focused on image-based AI processing, the AI HAT+ 2 comes with onboard RAM and can run small gen AI models like Llama 3.2 and DeepSeek-R1-Distill, along with a series of Qwen models. You can train and fine-tune AI models using the device as well.   $130 is lower than I expected, based on the price for the other Hailo stuff. More on the Hailo H10 . Could probably put together a “play around with” setup for ~$200, a bit less than I ultimately have invested...

Over the weekend, I had the occasion to search Google for so much information, trying to troubleshoot and bring back online my main server . Not only were all of the AI answers just  wrong  (often subtly, but still, wrong), Safari on the iPad does not play well with the AI responses (the Google page would refresh a few times before the browser gave up). I had to append -AI to all of my queries, to suppress the AI output . Sigh. I did a plain language search on Google again today and the AI response was ... Humorous, if the implications weren’t so dire. The MH-100 QMNT has a  quilted  maple top (hence the “QM”) (nothing “quantum” about it), and the NT stands for “no tremolo” – it’s a hard tail with a Tune-o-Matic bridge. But that’s not the biggest issue. The MH-100 definitely does  not  have a 3+3 tuner configuration (the Les Paul-style EC series does); it has a pretty standard inline 6 angled headstock:   MH-100QMNT Headstock I have more to write on t...

This has been bugging me for a minute. I remembered a site, 'raging.com,' that was a de-bloated AltaVista service, that took a page from the early Google playbook and offered fast, powerful, text-only searching. Trying to find it circa the 1990s via archive.org found captures of Raging Bull (“Leading the Investor Revolution”), located at www.raging.com , through at least May 11, 2000. And since I also remember visiting that site a lot during those heady dot-com bubble days, I was convinced I was mis-remembering the name of the AltaVista search offering. But, nope, my memory is (at least in this respect) intact, and it wasn’t a fever dream, confirmed when I found this Slashdot comment . As early as May 19, 2000, raging.com brought you to raging search from altavista : raging.com search engine home page By August 15, 2000, it was a redirect to ragingsearch.altavista.com , but still branded “raging.” Looks like that survived until at least early 2005 , but by February 5th of that ...

A cassette futurism AI (kinda) ... But you knew that. Well, you should have known that. If you didn’t, here’s a few articles that underscore why you shouldn’t: AI chatbots fail at accurate news, major study reveals AI assistants make widespread errors about the news, new research shows AI Not A Reliable Source Of News, EU Media Study Says ... and of course the study itself:  News Integrity in AI Assistants  That said, I’ve been playing with my cheap AF AI workstation  (and low key coveting the $4,000 Nvidia Spark ), and I do see places, narrow places, where LLM type maths can be a boon. But, like, the other day I got a wall of text from a friend (who described it as “a synthesis based on a mass of affirmative data” 🤣) that just got things ... wrong. And a few days before that I’d asked some AI tools to summarize some expedited procedural stuff in a very niche practice area, and each one gave me the same – very wrong – information. When I rolled up my sleeves and manuall...

I had an old (circa 2012?) Hackintosh, which started life as a Core i5 3570K Processor (3.4 GHz 4 Core), installed on a Gigabyte GA-B75M-D3P  main board (LGA 1155 socket, B75 chipset). Basically, a lightly tweaked version of this moarfish build . The Corsair Carbide 300R case (or, rather, the plastic drive rails) were showing their age, the power supply had a tendency to run extremely - worryingly - loudly, and so it was just sort of collecting dust. A lot of dust (see below). I had picked up a new Cooler Master case and EVGA power supply to rebuild it, but never got around to it. Instead, I eventually moved to a maxed out (1TB SSD, 16GB RAM, Core i7-3720QM CPU @ 2.60GHz) 2012 Mac mini server, running the latest macOS through OCLP. Now that I’ve finally decided to pull the trigger and start getting my hands dirty with “AI,” after flopping around for a minute looking at various options (eGPU chassis hooked up to a 2020 MacBook Air running Linux? $249 NVIDIA Jetson Orin Nano Super De...