Raphaël Pinson

Raphaël Pinson

@raphink.info

@cilium.io Alchemist @isovalent at Cisco | CNCF Ambassador | I'm the Oops in DevOops ∞ | Family Historian 🌳 | Autism Awareness | Tweets Mine | @raphink@hachyderm.io

3K followers1.1K following458 posts

When neurotypical people read someone's expression, it happens automatically: parallel processing, fast, effortless. Like a GPU rendering a scene. #autism #actuallyAutistc #neurodiversity 🧵

When an autistic brain does it, I run a different process: observe, decode, match, select, output. Sequential, non-automatic, constructed from scratch. The result looks similar, but the cost doesn't.

This is why masking is exhausting: not because autistic people are bad at social interaction, but because they're running a CPU process where everyone else has a GPU. Same output. Different architecture.

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: 👇 labs.isovalent.com #cloudnative #kubernetes #networking #security

AI video generation destroys color precision. Color detection → hollow eyes. Chroma key → magenta speckles everywhere. The fix: stamp the original PNG's alpha channel onto every frame. No detection, no approximation. The mask was already there. New post 👇 dev.to/raphink/clea...

dev.toClean Edges: Using a PNG Alpha Mask on AI-Generated AnimationsHow I solved the background removal problem for AI-generated animated WebP tiles, and why color detection and chroma key both fail.

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. 🧵

View full thread

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. 🧵

Chat threads are the wrong tool for AI image generation. 4 years of drift taught me that. So I built panelgen: declarative, idempotent image generation from a YAML spec. 🚀 brew tap raphink/tap && brew install panelgen Full story 👇 dev.to/raphink/four... #ai #generativeAI #opensource #automation

Two kinds of anxiety — completely different sources. 🧠 Psychological (top-down): all sensors green, but the cortex generates a loop anyway. "What if..." Cognitive strategies can help here. #autism #actuallyAutistic #neurodiversity 🧵

View full thread

When the hardware runs at permanent 97% alert status, you don't crash from worry. You crash from biological burnout. Don't confuse a hardware alert for a software bug.

Two kinds of anxiety — completely different sources. 🧠 Psychological (top-down): all sensors green, but the cortex generates a loop anyway. "What if..." Cognitive strategies can help here. #autism #actuallyAutistic #neurodiversity 🧵

🚦 Physiological (bottom-up): subcortical sensors are natively running above the normal alert baseline — even when everything is safe. Cognitive regulation doesn't fix it. The loop didn't start in the cortex.

So I watched Bicentennial Man yesterday evening. If it wasn't for the robot disguise, I'd be sure Robin Williams was trying to portray an #actuallyAutistic person -- and probably doing a better job at it than if it had been intentional. The whole time, I was like -- hey, that me 🤔 🧵

That got me to think (especially after my last post on AI and humanity) about how defining humanity by (a certain form of, most likely NT-shaped) empathy would exclude most #neurodiversity people from that definition.

Everyone is talking about hallucination. That's the wrong diagnosis. Hallucination isn't a bug — it's the mechanism. Turn it off and you don't have a more reliable model, you have a lookup table. The real problem: non-determinism misallocated. dev.to/raphink/ai-d...

dev.toAI Doesn't Hallucinate. Your Architecture Does.Hallucination isn't a bug in LLMs — it's the mechanism. The real problem is misallocating non-determinism, and why "SKILLS.md is enough" is exactly backwards.

Defining Autism: The Shadow and the Object The DSM defines autism by what clinicians can observe. But describing the shadow isn't the same as defining the object. Maybe we need to redefine it? dev.to/raphink/defi... #autism #autismawareness #neurodiversity #actuallyautistic

dev.toDefining Autism: The Shadow and the ObjectThe DSM defines autism by observable symptoms. But a definition and a diagnostic test are not the same thing. One bar has been cleared. The other hasn't. And mistaking the proxy for the definition has...

#GenAI is an amplification machine. Good ideas in → sharper. Bad ideas in → worse, more confidently. Nothing in → statistical word soup. The gate was never the tool. It was always the ideas. #AI #Writing #LLM dev.to/raphink/gen-...

dev.toGen AI is an Amplification MachineMost people misunderstand what generative AI actually does. It doesn't add quality — it multiplies what's already there. Which means the people it serves best are the ones who least needed it to begin...

What would it look like, in a world full of autistic people, to study and try to understand how allistic (non-autistic) people function? Attempting to explore this (in a cheeky way) in this new article: dev.to/raphink/retu... #autism #actuallyAutistic #neurodiversity

dev.toReturn to the Planet of the AutisticsA field journal from an imaginary researcher studying a puzzling minority neurological condition called allism, which affects approximately 1% of the population.

Why has autism persisted across all human populations, across cultures, and most likely across centuries? A hypothesis: not individual advantage, but group survival. dev.to/raphink/why-... #autism #autismAwareness #ActuallyAutistic #Neurodiversity

dev.toWhy autism hasn't disappeared — a hypothesisWhy has autism persisted across all human populations, across cultures, across centuries? A hypothesis: not because of individual advantage, but because groups that contained the profile were more rob...

My last post on #autism introduced a diagram intending to conceptualize Milton's double empathy problem at the core. Does that speak to you?

La maison où j'ai grandi est à vendre. Jardin avec mare et tritons, une trentaine d'espèces d'oiseaux, forêt à 5 min à pied, RER B à 10 min en bus, Paris-Saclay à 15 min en voiture. Si vous connaissez quelqu'un qui cherche dans ce secteur... 🌿 => maison-bures.fr

maison-bures.frMaison à vendre — Bures-sur-Yvette, Vallée de ChevreuseMaison 6 pièces, 105 m², jardin 489 m² avec bassin et potager. Forêt de Chevreuse à 5 min, RER B à 10 min en bus, Paris-Saclay à 15 min. 436 800 € FAI.

#Neurodiversity and the two layers of cognition What if the way you think and the output you produce are two completely separate things — and the gap between them has a cost? A post about internal process, expected output, and what autism adds to the picture. #autism dev.to/raphink/neur...

dev.toNeurodiversity and the two layers of cognitionWhat if the way you think and the output you produce are two completely separate things — and the gap between them has a cost? A post about internal process, expected output, and what autism adds to t...

I built an MCP server for meta-search in online Newspapers archives, and in a matter of one hour, Claude found 5 major documents I had never found before and documented them in Notion 🤯 #genealogy #ai