Raphaël Pinson

A few years ago, I realized my wife's condition had something in common with #autism and I thought I'd make a comic strip about it, to help understand both.

She has demyelination: her nervous system doesn't automatically coordinate her movements the way most people's does. Every step is conscious, planned. And some signals still arrive too late, no matter how much practice.

With autism, social and emotional signals don't get pre-filtered before reaching conscious awareness. What most people process automatically, we process deliberately, with a similar risk of arriving too late.

🧵

👉 https://raphink.info/autism

#actuallyautistic #neurodiversity

@autistics@fedigroups.social

Most people picture autism as a line from "a little autistic" to "very autistic."

That's not what the spectrum means.

#Autism is not a dial, but a different way of seeing.

🔗 raphink.info/autism

#actuallyautistic #neurodiversity @autistics@fedigroups.social

🧵

A couple of years ago, learners on our platform kept complaining they never got their badge. Some redid the lab a second time, even a third, hoping that would make it appear.

The badge had been issued every single time.

Instruqt sends a webhook when the lab finishes. The platform catches it, processes it, asks Credly to issue the credential. Credly emails the learner a claim link. Every piece of that chain worked. But the webhook is asynchronous — it can land after the exam screen is already gone. So learners finished, saw nothing, and assumed it broke.

I could have called this a delivery bug, but it wasn't. The system was functioning exactly as built. The gap was between what the system did and what the learner expected to happen next.

So I measured two things: how often people complained, and how often they redid a lab they'd already passed. Then I made the smallest change I could think of — one slide before the final exam, telling learners an email was coming.

Complaints fell sharply. The pointless repeat attempts mostly disappeared.

No API changed, no webhook became synchronous. One sentence, in the right place, fixed the flow.

A credential isn't done when it's issued. It's done when it's claimed, used, and — if the learner wants — shared. My job doesn't end where my component's success metric does.

Read the full story, and how Goldratt's The Goal (not The Phoenix Project, the original) shapes how I think about technical education as a system, and infrastructure as a whole:

https://dev.to/raphink/the-badge-was-issued-but-the-goal-wasnt-achieved-41l3

Time to rethink my portfolio page!

https://raphink.info/

Generating a video from an AI model runs the risk of losing something precious: color precision.

I was building animated tiles for a new UI — isometric city-builder scenes, one per learning zone, that come alive on hover. Each tile starts as a static PNG, then gets animated as a short looping video.

The obvious next step: remove the background, serve as an animated WebP with transparent background.

Except color detection failed: the black background bled into the subject, creating hollow eyes and transparent rooftops. Chroma key also failed: the model interpolated the magenta into a cloud of pinks scattered inside the subject. Every fix made the edges worse.

Then I realized: I already had a clean mask. The original PNG had a perfect alpha channel. And since I enforced no camera movement, every frame of the animation fit within that same boundary.

So instead of detecting the background, I extracted the alpha from the PNG and stamped it onto every frame. No detection, no approximation. Exact, deterministic, scriptable.

Don't use a probabilistic tool when a deterministic one can do the job better and faster.

New post in the Misallocated Machine series 👇
https://dev.to/raphink/clean-edges-using-a-png-alpha-mask-on-ai-generated-animations-ifb

#animation #webdev #ai #tooling

Post 12 in my #autism Awareness Series, on autism working as a catalyst for other conditions:

https://dev.to/raphink/autism-as-a-catalyst-322e

#autismAwareness #neurodiversity #actuallyAutistic

@autistics@fedigroups.social

It's never been more fun to learn about Cilium, Tetragon, and eBPF!

The Isovalent Labs Platform has a new look, with improved learning journeys and progress tracking.

Start (or continue) your Cilium learning adventure:

👇

https://labs.isovalent.com

#cloudnative #kubernetes #networking #security

Post 10 in my #autism awareness series, on autism and the "genius" effect.

https://dev.to/raphink/autism-and-the-genius-effect-k86

#autismawareness #actuallyautistic #neurodiversity

Debirdify is struggling today.

Use Fedifinder instead: https://fedifinder.glitch.me/success.html

And tell your friends to add their Mastodon handle to their Twitter profile!

Would you trust a sci-fi author to program critical AI systems for humanity? No? Yet, that's what we've been doing.

LLMs are autocomplete engines trained on human culture. Sci-fi authors spent decades writing how robots fail. That's the training data. That's the completion.

This isn't a bug, it's the design working exactly as intended.

New post in The Misallocated Machine series 🤖

#ai #llm #scifi

🧵

👉 https://dev.to/raphink/we-let-sci-fi-authors-code-ai-for-us-1j1e

@adelinej@piaille.fr Merci!