Mike Goatly

Mike Goatly

@goatly.net

Professional software developer. Guitarist and pretend musician. You can expect me to post about Visual Studio, VSCode, C#, .NET, TypeScript, React, Avalonia, Raspberry PI, Azure, Kubernetes and more.

35 followers117 following86 posts

Notice the em-dash there. He used Claude to write this I'm >10% sure of it. (Why? It's ~50 words!)

Carl Quintanilla@carlquintanilla.bsky.social

ANTHROPIC EXEC: “.. I personally think it is >10% within the next decade.”

Mike Goatly reposted

I really wonder about the consequences of an entire generation having a Joker style origin story. High school years impactsd ruined by COVID only to graduate college into a job market where companies brag about preferring Claude to hiring you.

Getting close to launching the new @avaloniaui.net based version of #Chordle on Windows. I've spent time building a headless test suite for common scenarios so I can safely build on it in future - but there's no reason why a headless test suite can't let you visualise what's actually going on!

There's a core testing library I'll probably open source at some point - it borrows quite heavily from React Testing Library with reusable Get*, Find* and Query* methods.

I've been working on a rebuild of #chordle for a couple of years now. Spare time projects can be brutal, and Chordle is a big ol' project, but I'm really happy with how the new editor is working out - drag and drop of chords in both #chordpro and chords above text! #avaloniaui

My new job has required I use Google's Firebase, and in particular Firestore. I've been surprised by the memory consumption of the official emulator, and wondered if I could hack together in .NET that meeds our needs. I still don't know, but this is the result so far: github.com/mikegoatly/F...

github.comGitHub - mikegoatly/FirestoreSharp: An open source .NET emulator for Google Cloud FirestoreAn open source .NET emulator for Google Cloud Firestore - mikegoatly/FirestoreSharp

Built a diagnostics dashboard so I can finally debug the raw ML probabilities against the decoded notes Next up: a mountain of work on note decoder tuning, quantization and score conversion. It's very possible I underestimated the size of this project!

I've just taken an early iteration of the ML model, quantized it to int8 (it's under 10Mb now!) and updated the audio transcription pipeline in StaffSharp to use it. It’s already successfully picking out the notes of my poor quality C-scale recording and converting them to sheet music!

For comparison, here's what it looked like with the pYIN algorithmic approach. Notes aligned better (I've not done any real work tuning the polyphonic quantizer yet) but notice it had real trouble with the starting middle-C it always got it an octave out.

So after training the StaffSharp model for another 9 hours, things are looking good - I can see that it's really starting to pick notes out, even that rapidly ascending sequence to the right

View full thread

The training is done in python, but I wrote the data prep logic in C# to guarantee that during inference, the audio data is calculated exactly the same way (no DSP mismatches!).

The plan is to slot this into my existing pipeline to replace pYIN, so all my downstream score transformation logic should still just work.

This whole project has been such a massive rabbit hole, but so interesting. I've learned that state-of-the-art piano transcription relies on ML, so I've started training my own model on the Maestro v3 dataset. It's really exciting to come back every couple of hours and see the accuracy improving.

Well this isn't bad for a vibe coded weekend project. Taking audio samples, processing them and transforming them into music notation, midi or ABC notation. It's a long way from perfect, but it's been a lot of fun and I've learned loads.

View full thread

This is an example of something that I've wanted to try for *years* but have never found the time. It was an especially full weekend and still to be able to get this far is remarkable.

Well this isn't bad for a vibe coded weekend project. Taking audio samples, processing them and transforming them into music notation, midi or ABC notation. It's a long way from perfect, but it's been a lot of fun and I've learned loads.

There are components for DSP (separate implementations for monophonic and polyphonic), beat detection, tempo detection, a CLI, a UI. There's a common internal representation, and modules to export in different formats. And tonnes of unit and integration tests.

Mike Goatly reposted

Replying to @gergely.pragmaticengineer.com

*or throw out code review for something else. Like prod observaility, automated testing etc. Standout engineers still write concise code with AI assistance as well… but it’s just so tempting to let it rip and push “sloppy code” that solves stuff but is verbose + less maintainable?

Mike Goatly reposted

A update: Per her last update on GFM, Rebecca's doctors have determined no further treatments are possible, and they will be ceasing life support later tonight. Please make time tonight, to tell her thank you, and that she is loved. It doesn't have to be much.

Retro Computers@retrocomps.bsky.social

The Bard's Tale III. Wasteland. Wolfenstein 3D. Jazz Jackrabbit 2. Out Of This World. Baldur's Gate II. Neuromancer. Battle Chess. Rock & Roll Racing. And, let us not forget: Super Noah's Ark 3D! Rebecca's work is everywhere in this industry, and she deserves our support and our love.

Exciting news from the Avalonia team - I hope to be kicking the tyres of this in v12...

Avalonia UI@avaloniaui.net

We're experimenting with additional rendering backends for Avalonia. SkiaSharp remains the default (and will for years), but we're exploring what GPU-first rendering could bring to the platform. Read about our early experiments: avaloniaui.net/blog/the-fu...

Mike Goatly reposted

OpenAI is being sued by the family of a teen who died by suicide after months of confiding in ChatGPT. The suit claims GPT-4o “validated his most harmful thoughts” and lacked safeguards. I found this screenshot of one of their chats most upsetting as a parent of a teen.

This reminds me of the "torturing furbies" phenomenon. Some people have a bias to cruelty and this is another way to spot it.

This post is unavailable.