@tsevis

Computational artist and designer. Algorithmic mosaics: images built from thousands of pieces. I build my own tools (Mozaix). https://xitter.cf/t.co/b9XlHAl6ET

 Paphos, Cyprus
tsevis.com
Joined February 2008
  • Tweets45.6K
  • Following1.6K
  • Followers5K
African Bricks: portraits built from thousands of fragments, inspired by the brick-and-colour patchwork of townships like Soweto, where colour is a declaration, not decoration. From across the room, a face. Up close, a wall of colour. My most collected series.
Algorithmic mosaic portrait of a young African woman in a colourful headwrap, looking straight at the viewer. Her face and wrap are built from hundreds of tiles of African textile patterns in terracotta, rose, indigo and gold. "Will You?" from African Bricks by Charis Tsevis.

ALT Algorithmic mosaic portrait of a young African woman in a colourful headwrap, looking straight at the viewer. Her face and wrap are built from hundreds of tiles of African textile patterns in terracotta, rose, indigo and gold. "Will You?" from African Bricks by Charis Tsevis.

Close-up of the same mosaic around one eye: every tile is a different printed pattern, zigzags, dots, diamonds and stripes in browns, pinks and terracotta, separated by thin dark lines.

ALT Close-up of the same mosaic around one eye: every tile is a different printed pattern, zigzags, dots, diamonds and stripes in browns, pinks and terracotta, separated by thin dark lines.

2
1
108
1966, Bell Labs. Ken Knowlton and Leon Harmon digitise a photograph of a reclining figure and rebuild it from small black-and-white symbols. Each symbol has a measurable density: darker areas get denser marks. No photographs inside it yet. But the whole principle is there: many small marks, one large image. Sixty years of mosaic algorithms, read by someone who makes them: tsevis.com/blog/history-of-m…
A 1966 Bell Labs image by Knowlton and Harmon: a reclining figure rebuilt from small black-and-white symbols of different densities, with a magnified detail of the symbols at left.

ALT A 1966 Bell Labs image by Knowlton and Harmon: a reclining figure rebuilt from small black-and-white symbols of different densities, with a magnified detail of the symbols at left.

6
Teaching a machine to read old pages. It strips away the paper, leaving only the ink. Then it finds the lines and cuts them into words and letters. Each piece becomes a tessera for a new mosaic. Two ways of reading: one follows the text faithfully; the other sees the whole page, including the ornament. #computervision #workinprogress
1
53
more replies
The opening page of the Inferno (Canto I), from Nicolaus Laurentii’s edition with Cristoforo Landino’s commentary. The scene at the bottom is a copper engraving after a design attributed to Botticelli. The floral border and the large initial “N” were painted by hand. Dante’s text is the narrow block on the left. Landino’s commentary is the wide block around it. Source: commons.wikimedia.org/wiki/F…
1
15
This is the oldest complete manuscript of the Hebrew Bible. It is dated to 1008–1009 CE and was written in Cairo by the scribe Samuel ben Jacob. The National Library of Russia in St. Petersburg holds it (shelfmark Firkovich B 19 A). It is the basis of the main modern scholarly editions of the Hebrew Bible, such as Biblia Hebraica Stuttgartensia. Public domain. Source: commons.wikimedia.org/wiki/F…
10
Charis Tsevis retweeted
Tanishq (tk)
@tanishqkOct 23
my new favorite way of navigating user journeys
Tanishq (tk)
@tanishqkOct 23
share modals are complicated
6
12
2
319
23,482
The best part of this GenAI era is that when you need a tool, you can build it on your own. This weekend I needed an ebook converter for my Mac. The usual answer is Calibre, which is a whole library manager when all I wanted was to turn one format into another. So I built one. It’s free, open-source, and notarized by Apple: ts between PDF, EPUB, MOBI, DjVu, FB2, TXT and more, • reads scanned books with OCR, including Greek and polytonic Greek • converts whole folders at once, keeping their structure • lets you set the title, author and cover before converting • needs nothing else installed: every tool is inside the app The best part of this GenAI era is that you can build a tool yourself when you need one. #opensource #MACOS
The biblioteca welcome screen. A photo of a crowded bookshelf full of books about Apple, Steve Jobs and Silicon Valley carries the app’s name, an icon of a human figure with outstretched arms, the tagline “Converts books between PDF, EPUB, DjVu and more”, and “Version 0.3.0”. Below it the text explains that biblioteca reads PDF, EPUB, MOBI, DjVu, FB2, Word and more; writes PDF, EPUB, DjVu and text; and reads scanned pages with OCR. Everything it needs is inside the app, nothing leaves the Mac, and originals are never changed. The footer reads “Created by Charis Tsevis, with the help of Claude Code”, with links to tsevis.com and github.com/tsevis and a blue Continue button.

ALT The biblioteca welcome screen. A photo of a crowded bookshelf full of books about Apple, Steve Jobs and Silicon Valley carries the app’s name, an icon of a human figure with outstretched arms, the tagline “Converts books between PDF, EPUB, DjVu and more”, and “Version 0.3.0”. Below it the text explains that biblioteca reads PDF, EPUB, MOBI, DjVu, FB2, Word and more; writes PDF, EPUB, DjVu and text; and reads scanned pages with OCR. Everything it needs is inside the app, nothing leaves the Mac, and originals are never changed. The footer reads “Created by Charis Tsevis, with the help of Claude Code”, with links to tsevis.com and github.com/tsevis and a blue Continue button.

The biblioteca app window on macOS with the Batch tab selected. Below a search field reading “Search books and folders” and a count of 4 books, the list shows four PDF files: yokAcikBilim_464640.pdf, 2603.14505v1.pdf, 2608.03575v1.pdf and 2608.01195v1.pdf. Under each name is its conversion route, “PDF → EPUB · Tesseract → biblioteca”, and a tag icon for editing title, author and cover sits on the right. A bar along the bottom reads “Convert to EPUB”, with the output folder set to Desktop, “OCR scans” and “Skip converted” ticked, and a blue Convert button.

ALT The biblioteca app window on macOS with the Batch tab selected. Below a search field reading “Search books and folders” and a count of 4 books, the list shows four PDF files: yokAcikBilim_464640.pdf, 2603.14505v1.pdf, 2608.03575v1.pdf and 2608.01195v1.pdf. Under each name is its conversion route, “PDF → EPUB · Tesseract → biblioteca”, and a tag icon for editing title, author and cover sits on the right. A bar along the bottom reads “Convert to EPUB”, with the output folder set to Desktop, “OCR scans” and “Skip converted” ticked, and a blue Convert button.

1
1
116
Most large local models assume you can fit their weights in memory. Slotstream tackles the opposite case: what if your Mac can’t? It runs a roughly 105 GB, 125B parameter mixture-of-experts model on Apple Silicon Macs with 16–64 GB of memory, reading weights from SSD and keeping frequently used experts in RAM. A genuinely interesting project for anyone following local AI. github.com/carloslfu/slotstr…
Qwen AI Fans
@QwenAIFansOct 25
💥 You don’t need a 128 GB Mac for a 105 GB Qwen model anymore. Slotstream streams Qwen3.8-Flash-Next from your SSD — runs on 16–64 GB Apple Silicon Macs. One Swift binary, no Python, offline after download. Their 48 GB M5 Pro (33 GB budget): 20.91–23.69 tok/s across short / long / coding prompts. Caveat: Slotstream’s own Oct 5 benches — your Mac/SSD will differ. via GitHub 👇 github.com/carloslfu/slotstr…
1
156
The best thing I ever learned about making images, I learned from watching people who were not trying to impress anyone. Children make no distinction between a tool and a toy. Hard to design for, and the most difficult audience worth designing for.
John Ternus
@johnternusOct 26
Had so much fun with the students of Summitt Elementary in Austin!
74
The series and signed editions: tsevis.com/shop/african-bric…

African Bricks — Shop — Tsevis Studio

tsevis.com
13
Chris Pappas – The Greek Portrait More than a century ago, Arthur Pappas left Greece for New Hampshire. In 1917, he and his cousin opened a small ice cream shop in Manchester. It grew into the Puritan Backroom, which the family still runs today. His great-grandson, @RepChrisPappas, is now running for the U.S. Senate. ,I built this portrait from the architecture of Greece, gathered from many places across the country: island houses and chapels, domes with crosses, arches, staircases, ornaments and classical columns, in five shades of blue and white. A piece of the land his family left, sent back to him. From Greece, with all my heart: Καλή επιτυχία, Chris! 🇬🇷🇺🇸
4
122
Six typographic mosaic portraits of @PeteButtigieg. Each one is 4,000+ pieces of slogans, state names and speech lines. From afar a face, up close a crowd of messages. Can you find your state? tsevis.com/portfolio/pete-bu… #petebuttigieg #algorithmicart #typography
1
1
132
AI is not a war, and it is not a person. It's software. My talk at Digital Agenda 2026 in Nicosia: what designers, artists and illustrators should actually fight for. Three pillars: calm, useful assistance and, above all, quality. Full talk (27 min): youtube.com/watch?v=HrVy_gvA…

The Quality Wars: What Creatives Must Fight For in the Age of AI | Charis Tsevis

Artificial intelligence is not a war, and it is not a person. It is...

youtube.com
103
Charis Tsevis retweeted
Unsloth AI
@UnslothAIOct 15
Meta releases Muse Spark Open, a new SOTA open model. Run on a Mac Studio via Dynamic 2-bit Unsloth GGUFs. Muse-Spark-Open-816B-A42B supports text, images, video and audio input with 1M context. GGUF: huggingface.co/unsloth/Muse-… Guide: unsloth.ai/docs/models/muse-…
Mark Zuckerberg
@finkdOct 15
Muse Spark Open -- weights releasing today -- runs great on new Mac Studios thanks to work with @johnternus and Apple.
35
76
16
617
34,329
I was invited to talk about the "AI wars". I opened by saying there is no war. AI is software. The real question for creatives: is my work getting richer, or just faster? My Digital Agenda 2026 talk, written up: tsevis.com/blog/the-quality-…
1
2
108
Charis Tsevis retweeted
Gary Marcus
@GaryMarcusOct 12
shh, don’t tell the grifters
Dagfinn Reiersøl | Superintelligence@dagfinnrOct 12
Replying to @GaryMarcus
Not being able to follow simple instructions is a cognitive deficit relative to humans.
3
3
49
8,721
Sixty years of photographic mosaic algorithms, read by someone who has made mosaics since the 80s. From Bell Labs 1966 to diffusion models. #algorithmicmosaic tsevis.com/blog/history-of-m…
1
89
Charis Tsevis retweeted
Gary Marcus
@GaryMarcusSep 6
“The expected value of a fully LLM-generated, non-human-reviewed patch is a net-negative by a considerable margin.”
Hedgie
@HedgieMarketsSep 6
🦔AI can only fix security vulnerabilities 26% of the time. Researchers at 1Password ran over 6,000 AI-generated patches using Claude and ChatGPT against real vulnerabilities. Half the time the AI failed to fix the original bug. 4.5% of the time it created a brand new vulnerability that didn't exist before. And the researchers found that checking an AI-generated security patch takes more effort than just writing the fix yourself. Their conclusion was blunt. "The expected value of a fully LLM-generated, non-human-reviewed patch is a net-negative by a considerable margin." My Take OpenAI launched a program this summer called "Patch the Planet" where AI finds bugs and generates the fixes. These researchers ran 270 patches against one of those same bugs. Zero clean fixes. Not one. Every patch that fixed the original problem created a new vulnerability in the process. The partner that submitted a fix through OpenAI's program produced what the researchers classified as the worst possible outcome, it didn't fully fix the bug and it introduced a new exploit on top of it. Here's what this means if you don't write code for a living. Companies are using these AI tools to patch the software that runs your bank, your hospital, your phone. The pitch has been "AI finds and fixes security holes faster than humans." This study says the AI fix is four times more likely to be broken than working, and a third of the time it recreates the exact same mistakes human programmers already made. Reviewing the AI's work takes longer than doing it yourself. So the speed advantage disappears the moment you try to verify the output, which most companies won't do because the entire point was to move faster. I think we're going to see major breaches traced back to AI-generated patches that nobody checked, and the companies that shipped them are going to blame the tool instead of the decision to trust it. Hedgie🤗 Study: 1password.com/files/resource…
23
87
4
581
34,894
Hello, hello! – at Cupertino, CA
1
148
Load more