Julian
@shortcause.bsky.social
I discuss ideas & solution concepts, open for improvement. My focus is AI research and safety.
Reposted by Julian
For a detailed, nuanced exploration of the inner workings of neural networks, look no further than @shortcause.bsky.social's new deep dive on mechanistic interpretability.
Mechanistic Interpretability: Peeking Inside an LLM | Towards Data Science
Are the human-like cognitive abilities of LLMs real or fake? How does information travel through the neural network? Is there hidden knowledge inside an LLM?
towardsdatascience.com
February 7, 2026 at 7:18 PM
For a detailed, nuanced exploration of the inner workings of neural networks, look no further than @shortcause.bsky.social's new deep dive on mechanistic interpretability.
In my latest article, I offer a gentle introduction to mechanistic interpretability with LLMs, explaining how researchers dissect their neural networks. I also discuss the most amazing research results from the past 5 years! towardsdatascience.com/mechanistic-... @towardsdatascience.com
Mechanistic Interpretability: Peeking Inside an LLM | Towards Data Science
Are the human-like cognitive abilities of LLMs real or fake? How does information travel through the neural network? Is there hidden knowledge inside an LLM?
towardsdatascience.com
February 5, 2026 at 3:54 PM
In my latest article, I offer a gentle introduction to mechanistic interpretability with LLMs, explaining how researchers dissect their neural networks. I also discuss the most amazing research results from the past 5 years! towardsdatascience.com/mechanistic-... @towardsdatascience.com
More critique, more philosophy, more popular science. Looking for paths to a more human, more trustworthy, safe AI -- this will be the topic of my new German-language Substack: open.substack.com/pub/shortcau...
Hallo Substack!
Mein Ort, um einfach zu schreiben: Deutsche Essays, Kommentare und Gedanken zum Thema KI und KI-Sicherheit.
open.substack.com
January 25, 2026 at 10:07 PM
More critique, more philosophy, more popular science. Looking for paths to a more human, more trustworthy, safe AI -- this will be the topic of my new German-language Substack: open.substack.com/pub/shortcau...
www.science.org/doi/10.1126/...
This paper is such a smart summary of many relevant thoughts and a really nice read. Topics include
- an intro to functional computationalism
- the concept of gradual consciousness, also popularized by Hofstadter
..
This paper is such a smart summary of many relevant thoughts and a really nice read. Topics include
- an intro to functional computationalism
- the concept of gradual consciousness, also popularized by Hofstadter
..
Illusions of AI consciousness
The belief that AI is conscious is not without risk
www.science.org
September 14, 2025 at 4:37 PM
www.science.org/doi/10.1126/...
This paper is such a smart summary of many relevant thoughts and a really nice read. Topics include
- an intro to functional computationalism
- the concept of gradual consciousness, also popularized by Hofstadter
..
This paper is such a smart summary of many relevant thoughts and a really nice read. Topics include
- an intro to functional computationalism
- the concept of gradual consciousness, also popularized by Hofstadter
..
Reposted by Julian
Are we underestimating LLMs? 🤔 @shortcause.bsky.social's new article dives into why LLMs are more than just "next-word predictors," exploring evidence for creativity, internal world models, and even emotional associations. It's time to challenge our assumptions about AI intelligence.
Are You Being Unfair to LLMs? | Towards Data Science
They may deserve better.
towardsdatascience.com
July 11, 2025 at 10:32 PM
Are we underestimating LLMs? 🤔 @shortcause.bsky.social's new article dives into why LLMs are more than just "next-word predictors," exploring evidence for creativity, internal world models, and even emotional associations. It's time to challenge our assumptions about AI intelligence.
LLMs are often dismissed in everyday conversations for reasons that simply aren't true. Just a word predictor? No understanding, creativity, emotion, consciousness; if you ever heard or used these, my newest article is for you.
towardsdatascience.com/are-you-bein... @towardsdatascience.com
towardsdatascience.com/are-you-bein... @towardsdatascience.com
Are You Being Unfair to LLMs? | Towards Data Science
They may deserve better.
towardsdatascience.com
July 11, 2025 at 7:47 PM
LLMs are often dismissed in everyday conversations for reasons that simply aren't true. Just a word predictor? No understanding, creativity, emotion, consciousness; if you ever heard or used these, my newest article is for you.
towardsdatascience.com/are-you-bein... @towardsdatascience.com
towardsdatascience.com/are-you-bein... @towardsdatascience.com
My new article explores a topic I had only hinted at in the previous one: Applying OpenEvolve for text generation, that continuously evolves and follows different ideation paths. @towardsdatascience.com
Beyond Code Generation: Continuously Evolve Text with LLMs | Towards Data Science
Long-running content evolution and an introduction to result analysis
towardsdatascience.com
June 22, 2025 at 7:45 PM
My new article explores a topic I had only hinted at in the previous one: Applying OpenEvolve for text generation, that continuously evolves and follows different ideation paths. @towardsdatascience.com
Reposted by Julian
What if content generation wasn’t a single prompt, but an evolutionary process? @shortcause.bsky.social's new article shows how to continuously improve non-code text using OpenEvolve, with LLMs acting as both creators and critics.
Beyond Code Generation: Continuously Evolve Text with LLMs | Towards Data Science
Long-running content evolution and an introduction to result analysis
towardsdatascience.com
June 19, 2025 at 6:21 PM
What if content generation wasn’t a single prompt, but an evolutionary process? @shortcause.bsky.social's new article shows how to continuously improve non-code text using OpenEvolve, with LLMs acting as both creators and critics.
Reposted by Julian
Want to build AI that intelligently refines code? Julian Mendel's guide explores AlphaEvolve's core concepts and walks you through setting up OpenEvolve to experiment with evolutionary algorithms for program optimization.
Google’s AlphaEvolve: Getting Started with Evolutionary Coding Agents | Towards Data Science
Introduction AlphaEvolve [1] is a promising new coding agent by Google’s DeepMind. Let’s look at what it is and why it is generating hype. Much of the Google paper is on the claim that AlphaEvolve is…
towardsdatascience.com
May 22, 2025 at 11:04 PM
Want to build AI that intelligently refines code? Julian Mendel's guide explores AlphaEvolve's core concepts and walks you through setting up OpenEvolve to experiment with evolutionary algorithms for program optimization.
My first TDS article just got published:
"Google’s AlphaEvolve: Getting Started with Evolutionary Coding Agents"
towardsdatascience.com/googles-alph...
@towardsdatascience.com
"Google’s AlphaEvolve: Getting Started with Evolutionary Coding Agents"
towardsdatascience.com/googles-alph...
@towardsdatascience.com
towardsdatascience.com
May 22, 2025 at 9:16 PM
My first TDS article just got published:
"Google’s AlphaEvolve: Getting Started with Evolutionary Coding Agents"
towardsdatascience.com/googles-alph...
@towardsdatascience.com
"Google’s AlphaEvolve: Getting Started with Evolutionary Coding Agents"
towardsdatascience.com/googles-alph...
@towardsdatascience.com