Charis Tsevis

For the Computer Vision aficionados out there: Depth Anything 3 is here! ByteDance's latest model is way faster, more accurate, and more versatile than v2 or Meta's VGGT. A game-changer for computer vision, 3D reconstruction, and AR
Github: github.com/ByteDance-Se... ArXiv: arxiv.org/abs/2511.10647

Charis Tsevis - Caltech Magazine Cover Wins a UCDA Design Award Visual designer living on the sunny side of the planet. Photomosaics, computergraphics, neofuturism, digital art and graphic design

The Fall 2024 Caltech Magazine cover I illustrated, mosaics of extinct species made from “isotopic tesserae”, just received a UCDA Design Award.
I don’t believe in awards, but I believe in academia. Honored to collaborate with Caltech’s scientists on this one.

tsevis.com/caltech-maga...

From sci-paper analysis to sophisticated software. With basic skills, I used ClaudeAI as a partner to build it. My story:
tsevis.com/how-i-learne...
#GenAI #ClaudeCode

Charis Tsevis - When the Pioneer Plaque Becomes NSFW Censorship, AI, and the War on Context

When the Pioneer Plaque Becomes NSFW
Censorship, AI, and the War on Context: tsevis.com/censorship-a...

YouTube video by Times Radio The 'AI bubble' and what happens when it bursts | The Story

I will say that I think we are absolutely in a bubble. Things are crazy right now. However, that does not mean that AI is not going to change everything sooner or later, but I think both of those things can be true.
— Danny Fortson, US W.Coast correspondent, The Sunday Times
youtu.be/hm28qfJ6yTY?...

We should approach this with hope, not fear, so that this significant change benefits everyone. #AI #EthicalAI

It’s a really important opinion piece, and coming from a leader like Suleyman makes it even more meaningful. People like him and everyone who cares about technology should speak up clearly about where AI is headed.

His main idea is that AI should remain a tool to help us, not something we mistake for human life. If we blur that line, we might focus on questions like whether machines deserve rights, rather than ensuring AI is safe and helpful for all.

New AI systems may soon look and sound so real that people could start believing they are conscious. But they are not. They don’t feel or think like we do.

We must build AI for people; not to be a person Personal site of Mustafa Suleyman, AI pioneer and author.

"We should build AI for people, not to be a person!" Mustafa Suleyman, who helped start DeepMind and now leads Microsoft AI, wrote an essay with that quote as its title. The subtitle warns, "Seemingly Conscious AI is Coming." mustafa-suleyman.ai/seemingly-co...

The best part? He didn’t even do it alone. He straight-up asked Claude for help brainstorming the attack. And Claude delivered! It’s wild: these models are so powerful, they can literally help you break yourself. Watch the full demo here. Just promise you’ll use this knowledge responsibly, right?

YouTube video by Prompt Engineering GPT-OSS Jailbreak: No Fine-Tuning, No Hacks—One Simple Trick

However, Muhammad from Prompt Engineering actually demonstrates how easily it is to hack OpenAI’s model. No fancy tricks, no secret exploits, just one clever tweak and poof, the guardrails are gone.
www.youtube.com/watch?v=QTGr...

OpenAI keeps delaying its first ‘open’ model, saying it needs to be made bulletproof. After watching this video, you’ll see why safety is a massive problem with today’s AI. Sure, you could read about jailbreaks in some dry research paper.
#AISafety #AIHack

DINOv3 DINOv3 scales self-supervised learning (SSL) for images to produce our strongest universal vision backbones, enabling breakthrough performance across diverse domains.

Learn about the DINOv3: ai.meta.com/dinov3/
Grab the model: github.com/facebookrese...
Read the paper: arxiv.org/abs/2508.10104

Plug in a tiny adapter, and it adapts to your domain — sketches, UI mockups, glitch art — no fine-tuning circus. It’s lighter than you’d expect, open-source, and finally makes self-supervised vision feel usable, not academic.
#ObjectDetection #ScemanticSegmentation

Why is it my new go-to? While CLIP (from OpenAI or others) requires text captions and specialist models that overfit to niche data, DINOv3 operates like pure visual intuition. #SSLforImages

Forget begging for labeled data or wrestling with five different models. This model learns purely from raw pixels, 1.7 billion of them, and provides a single, frozen backbone that handles tasks such as depth mapping, segmentation, and material estimation out of the box.
#SSL #AI

Computer Vision is probably my favorite discipline of IT.
If you are building apps that enable your computer to see (style transfer, object removal, smart masking, visual effects etc.) with DINOv3 has quietly changed the game.
#ComputerVision

Also use weighted prompts based on relevance.
Works across 6 tasks. No fine-tuning. No extra data.
Interesting step in using models more intelligently, not just bigger or trained more.
arxiv.org/html/2508.09...
#ComputerVision #ML

A new paper shows how Stable Diffusion can do visual in-context learning without any training.
They change how attention works during inference to let the model compare a query image with example prompts.

ComfyUI Wan2.2 Fun Native Support and LightX2V 4-Step LoRA Integration ComfyUI now supports Wan2.2 Fun Control and Wan2.2 Fun InP

ComfyUI Wan2.2 Fun Native Support and LightX2V 4-Step LoRA Integration #ComfyUY #GenAI
blog.comfy.org/p/comfyui-wa...

Made with custom algorithms + 15 years of African textile research. 4th edition in my series.
#DigitalArt #AfricanCulture

Two parts: 5 imaginary women (archetypal spirits) + 5 real icons (Tyla, Siya Kolisi, Thuli Madonsela). Each uses 1000+ textile patterns. #AlgorithmicArt #ComputationalArt

Charis Tsevis - The African Bricks 4: Dreams and Icons For more than fifteen years, I’ve been learning from Africa’s culture. On my first trip to Johannesburg or Tunis, I wasn’t just a visitor. I felt changed as soon as I arrived. Africa leaves a lasting ...

The African Bricks 4 is live. 10 digital portraits celebrating African culture through textile patterns. #DigitalArt #DigitalMosaic
tsevis.com/the-african-...

YouTube video by Fahd Mirza Fine-Tune GPT-OSS-20B on Your Own Dataset Locally: Step-by-Step Tutorial

Fine-Tune OpenAI's open source GPT-OSS-20B on Your Own Dataset Locally: Step-by-Step Tutorial
www.youtube.com/watch?v=LRvX...

YouTube video by Jo Van Eyck AI coding agents are useless on large codebases. Unless you do THIS.

Results:
30% faster respons
40% fewer tokens ($$$ saved)
Works with your existing IDE
www.youtube.com/watch?v=Uqfx...
P.S. Take a look at the comments section for more MCP servers and other suggestions.
#ClaudeAI #ClaudeCode

YouTube video by Jo Van Eyck AI coding agents are useless on large codebases. Unless you do THIS.

Make Claude Code 30x Faster on Huge Codebases (Open Source and Free Fix!)
Struggling with Claude Code on large legacy codebases? Jo Van Eyck’s new video shows two free tools that actually work:
Serena MCP – Optimizes file ops/caching
Refactor-MCP – Pre-processes code to slash token use.
#AICoders

YouTube video by Alex Ziskind NVIDIA users: QWEN3 is FREE, but you’ll pay double

Here's why “free” QWEN3 coder can end up costing NVIDIA users twice as much while Apple owners skate free.
www.youtube.com/watch?v=0BHB...
#Apple #AI #OpenSourceAI

Sometimes you need to trace backward to understand where you're going. The full story surprised even me.
You can read the piece on my blog: tsevis.com/when-machine...
#DigitalArt #DigitalComplexity

The things he taught me in his workshop now mix with my computer coding. Hands-on meets digital. #CreativeCode #CodeArt #DigitalMosaic