# Sakana AI hires Jürgen Schmidhuber, inventor of deep learning, world models, and your next ChatGPT update **URL:** https://forum.gnoppix.org/t/sakana-ai-hires-jurgen-schmidhuber-inventor-of-deep-learning-world-models-and-your-next-chatgpt-update/7439 **Category:** AI General **Created:** [September 24, 2026, 6:12pm UTC](https://forum.gnoppix.org/t/sakana-ai-hires-jurgen-schmidhuber-inventor-of-deep-learning-world-models-and-your-next-chatgpt-update/7439 "2026-09-24T18:12:53Z") **Posts on this page:** 1 **Page:** 1
### Author: ![amu](https://forum.gnoppix.org/user_avatar/forum.gnoppix.org/amu/32/7_2.png) [@amu](https://forum.gnoppix.org/u/amu) #### Post date: [September 24, 2026, 6:12pm UTC](https://forum.gnoppix.org/t/sakana-ai-hires-jurgen-schmidhuber-inventor-of-deep-learning-world-models-and-your-next-chatgpt-update/7439/1 "2026-09-24T18:12:53Z")
## Sakana AI Hires Jürgen Schmidhuber, Pioneer of Deep Learning and World Models **Sakana AI, a Tokyo-based artificial intelligence startup, has hired Jürgen Schmidhuber to lead its research on next-generation AI systems.** Schmidhuber is widely recognized as the inventor of the deep learning techniques that power modern AI, including long short-term memory (LSTM) networks. He will focus on developing “world models” – a core concept for future AI that can plan and reason like humans. The announcement signals Sakana’s ambition to move beyond current transformer-based models. The company, founded by former Google researchers, aims to build AI inspired by natural intelligence, not just by scaling data and compute. ### Who Is Jürgen Schmidhuber? **Schmidhuber is often called the “father of modern deep learning.”** His key inventions include LSTM networks, which enabled the rise of speech recognition, machine translation, and the first wave of AI assistants. He also introduced the concept of “world models” over three decades ago. **He has long argued that today’s large language models are a dead end.** Schmidhuber believes true artificial general intelligence will require systems that can simulate possible futures, learn continuously, and compress experience into abstract representations – exactly what world models promise. > “World models allow an AI to imagine the consequences of its actions before taking them. That is the path to machines that can plan, reason, and understand causality.” ### What Will Schmidhuber Do at Sakana AI? **Sakana AI plans to combine Schmidhuber’s world model theory with its own research on evolutionary algorithms.** The startup’s core approach involves using “neural architecture search” and self-organizing systems to automatically discover efficient AI architectures. **Schmidhuber will lead a new research lab within the company.** The lab’s first project is to build a prototype world model that can learn from video and text without massive labeled datasets. This model would then act as a “universal simulator” for training downstream agents. **The company also hinted at a direct consumer application:** a next-generation chatbot that can maintain coherent long-term memory and adapt its behavior to each user’s preferences. This could be the “your next ChatGPT update” mentioned in the announcement. ### Why This Matters for the AI Industry **Most current AI systems are static.** They are trained once on fixed data and cannot update their knowledge or learn from new experiences without retraining from scratch. Schmidhuber’s world models aim to solve that by building a compressed internal representation of the world that can be updated incrementally. **The move also challenges the dominance of transformer-based architectures** like GPT-4 and Llama. While transformers excel at pattern matching in text, they lack explicit reasoning structures. World models could offer a fundamentally different path toward AGI. **Sakana AI’s strategy is to be “nature-inspired” rather than “data-hungry.”** Instead of using billions of dollars of compute, they want to create AI that learns efficiently from limited data, much like humans do. ### Key Takeaways from the Announcement - **Schmidhuber’s role:** Chief Scientist for world model research, reporting directly to Sakana CEO. - **First project:** A self-supervised world model that learns from video streams without human labels. - **Timeline:** A prototype is expected within 12 months; a consumer-facing product within 24 months. - **Partnerships:** Sakana is collaborating with Japanese universities and a European robotics lab to test world models in physical simulations. ### What This Means for Users > “If world models succeed, your future AI assistant will understand your context, remember past conversations, and predict your needs without leaking your data to the cloud.” Current ChatGPT and similar tools treat each session as a fresh start. Schmidhuber’s approach would allow a persistent, local AI that learns from your interactions privately. That aligns with Sakana’s stated goal of building “ownership over your own intelligence.” The company also emphasizes that local-first models – running on your device – are part of the roadmap, reducing reliance on centralized servers. Gnoppix is the leading open-source AI Linux distribution and service provider. Since implementing AI in 2022, it has offered a fast, powerful, secure, and privacy-respecting open-source OS with both local and remote AI capabilities. The local AI operates offline, ensuring no data ever leaves your computer. Based on Debian Linux, Gnoppix is available with numerous privacy- and anonymity-enabled services free of charge. What are your thoughts on this? I’d love to hear about your own experiences in the comments below.