
Technology Advisor & Business Engineer 🇨🇭 Practical AI & Data Science insights for real work — not just demos. Zürich | #DataScience #AI #ML
Zürich CH
Joined April 2011
- Tweets53.1K
- Following2.7K
- Followers3.3K
- Likes8.9K
claudîus retweeted
Someone just dropped 1.4 million pixel art sprites on Hugging Face 😱👾
free to use, captioned, 16×16
huggingface.co/datasets/Anga…
claudîus retweeted
ChatGPT has now a big problem.
Researchers at Oxford exposed a massive threat to large language models.”
It’s called "Model Collapse."
And it explains why the internet is about to become a toxic waste dump for Large Language Models.
Right now, AI companies are scraping the entire web to train their next-generation models.
But there’s a massive problem.
The internet is no longer just human data. It is rapidly filling up with AI-generated text, images, and code.
What happens when an AI trains on data generated by another AI?
A new paper in Nature just gave us the answer. And it is terrifying.
When AI models are recursively trained on AI-generated data, they suffer from irreversible brain damage.
The researchers found that the model starts forgetting the "tails" of the original data distribution. The rare, weird, unique, and complex human nuances completely disappear.
It's like making a photocopy of a photocopy.
Every generation loses resolution. The output gets flatter, blander, and more generic, until the model completely collapses in on itself.
For years, we thought the biggest bottleneck for AI was compute, getting enough H100 GPUs from Nvidia.
This paper proves the real bottleneck is us.
If we keep flooding the internet with synthetic AI slop, we will poison the very data pool required to build smarter models.
The ultimate irony? We thought AI was going to make human creators obsolete.
Instead, this research proves that verified, genuine human data is about to become the most valuable commodity in the tech industry.
claudîus retweeted
I see real potential in tokenized real estate. I also know the property still needs to be acquired well, managed well and sold or refinanced well. A digital ownership layer cannot fix poor property economics. #RealEstate #RWA
claudîus retweeted
Seven sources of AI debt include data quality issues, model drift, and overly entitled AI agents. The common one: AI governance that focuses only on policies without implementation. Policies aren't guardrails.
#AI #AIgovernance
drive.starcio.com/2026/10/go…
claudîus retweeted
Weekend hack: HackerTube. I was annoyed with YouTube's recommendations so I had Claude filter down 43M comments over 20 years of HN to 6,454 videos that keep being recommended. *Significantly* better viewing experience, ~2 hours with Opus 5.5:
hackertube.org/
- Excellent recs without any logins (state is kept in local browser session)
- Thumbnail is a representative frame from middle of the video instead of the super duper A/B test hyper-optimized user uploads
- Underlying data from the comment analysis as one zip file, point your Claude at it and ask for your own version
- HackerTube Longs ✊
A bit more on the process:
I use Claude Code (desktop app for Mac), Opus 5.5 on Medium effort as my default. I only talk to Claude and have it do everything, the only "manual work" I helped with in tandem was in the browser (reminiscent of MenuGen): 1) I had to create some kind of API key from Google Cloud UI to query YouTube for video metadata, 2) I booted up a VPS box for serving and 3) I bought the domain name. Once the box was up I gave Claude the IP and told it update, secure, harden, deploy and optimize the code, and it did the whole thing.
Many awesome projects that would have normally taken weeks/months even can now be few-shotted for breakfast. What else are we not thinking of.
claudîus retweeted
Probability distributions may look like separate mathematical tools, but many are connected by a simple change in conditions.
A binomial distribution approaches the Poisson distribution when n grows large and p becomes small, while np = λ stays constant. The normal distribution emerges from other limits, linking countless random events to the familiar bell curve.
The surprising part is that probability is not a collection of isolated formulas. It is a network of relationships.
claudîus retweeted
Mathematics is usually divided into a few large families: foundations and logic, algebra, number theory, geometry and topology, analysis, discrete mathematics, and probability and statistics. Applied mathematics draws on all of them.
The divisions are not clean. Algebraic topology studies shapes with group theory, analytic number theory studies primes with complex analysis, and probability is formally built on measure theory. The Mathematics Subject Classification used by journals lists 63 top-level areas.
Any map like this is a choice about grouping rather than a fixed fact. Which branch would you add, or put somewhere else?
claudîus retweeted
The real bottleneck in agentic AI isn't the model—it's the handoff between systems.
In practice, most "autonomous" workflows still fail at the API boundary: incomplete context, brittle schemas, or missing audit trails. Before you scale agents, map every tool call like a data pipeline—inputs, outputs, failure modes, and ownership.
How are you designing those handoffs today?
#AI #AgenticAI #DataScience #ML #DataEngineering #EUAIAct
claudîus retweeted
Hybrid Machine Learning & Financial Econometrics: Building Volatility-Aware Dynamic Forecast Bands for #Bitcoin
🔗datageeek.com/2026/10/21/hyb…
ℹ️ @YahooFinance
#AI #DataScience #Rstats #Dataviz #FinViz #QuantitativeFinance #QauntTraders #CryptoTrading #BTCUSD $SPY $QQQ
claudîus retweeted
Just noticed my skills repo is at 50,000 stars and 2,7 million downloads, thanks!
More skills coming soon, you’ll like them.
github.com/emilkowalski/skil…
claudîus retweeted
holy sh*t, Karpathy just put an entire AI education on GitHub for free.
7 repos. 300K+ stars combined. basically a full AI degree without paying a dime.
follow this order. each repo takes you one step further:
1. micrograd -> what "learning" really means
a whole neural net engine you can read over coffee
> github.com/karpathy/microgra…
2. nn-zero-to-hero -> the lectures
the notebooks behind his course, from a single neuron up to GPT
> github.com/karpathy/nn-zero-…
3. minbpe -> how a model reads text
the tokenizer step every LLM starts with, written clean and minimal
> github.com/karpathy/minbpe
4. nanoGPT -> train a GPT yourself
the simplest codebase to train and fine-tune one
> github.com/karpathy/nanoGPT
5. llm.c -> pop the hood
the same training in raw C and CUDA, no PyTorch doing the magic for you
> github.com/karpathy/llm.c
6. nanochat -> your own ChatGPT, start to finish
training GPT-2 cost about $43,000 in 2019
here you can beat it for about $48
> github.com/karpathy/nanochat
7. autoresearch -> AI that does the research for you
the agent tweaks the code, trains for 5 minutes, keeps what worked
about 100 experiments while you sleep
> github.com/karpathy/autorese…
people pay thousands for courses that teach less than this
finish the list and you'll understand AI better than most people who post about it
and that’s just the beginning. here are 15 more Claude Code repos most developers don’t even know exist
claudîus retweeted
Someone on here clones Beehiv because they think it’s too expensive but then someone else clones that person’s SaaS because they think it’s too expensive. And then someone else clones this new person’s product and releases it for free. I can’t stop laughing.
claudîus retweeted
Replying to @JackEllis
Then all of them make a total of $300 dollars, realize how bad it is, give up and re-subscribe to Beehiv 😂
🚨BREAKING‼️:
🇨🇭 Die Schweiz in Minecraft
- Echte Berge. Echte Täler. Alles begehbar.
Ein ETH-Student namens Manifold hat die gesamte Schweiz im Massstab 1:1 in «Minecraft» nachgebaut. 🇨🇭⛏️
Ein Minecraft-Block ist ein Meter Schweiz.
👉🏼 Es sieht echt WOW 🤩 aus. 🏔️ ⛰️
Das Aletschgebiet ist gratis, die gesamte Schweiz kostet ab CHF 15.- und ist nur für Personen mit Wohnsitz in der Schweiz zugelassen.
Läuft auf: Windows, Mac, Linux
Downloaden kann man die Packs auf:
blocks.manifold3d.ch/
- Gerne teilen, wenn Du das auch cool findest.
SpaceXAI engineer, Lauren Tan dropped best GrokBot template
here is prompt that makes Engineering lead even better
Engineering Lead owns the entire loop:
decompose → run → chase CI → review → report
it makes GrokBot:
→ inspect the repo before delegating
→ give every agent a complete brief
→ keep going until the PR is healthy
engineering lead + this prompt = GrokBot that doesn’t just start agents
it owns the work until every check passes and the PR is merge-ready
save this, then give your Engineering Lead the upgrade
claudîus retweeted
bros. Shenzhen is something else.
it's my birthday. a robot just woke me up with a cake.
nobody ordered it. the hotel has a system: if your birthday falls during your stay, it auto-orders a cake on Meituan ( chinese uber eats) and a robot delivers it. no human involved
claudîus retweeted
From sketch to home, built in 3D with Claude Opus 5.5 using Three.js + TSL. The architecture evolves through four stages: first lines, massing, detail, and finished home.
claudîus retweeted
GPT-6 Astra + Unreal Engine 5.8 = real estate UX boost
I transformed 20 photos of a real property listing on OLX into a 3D interactive space you can walk inside.
For exterior, we used Google Maps panoramas with relevant coordinates
This elevates the customers' decision making process & boosts the sales funnel (without spending a thousands on just a walkable render)
Let me know if you want me to share the full process

































