Sakana AI announces in-house pretrained LLM series 'Sawara', first model due spring 2027
hardmaru · x · 2026-11-02
At its Tokyo symposium "From World Models to Real-World AI" (400+ attendees), Sakana AI announced it is moving into in-house pretraining. Chief Scientist Takuya Akiba unveiled the "Sakana Sawara" (tentative) LLM series, with the first small model slated for spring 2027. Chief Scientific Advisor Jürgen Schmidhuber — pioneer of pre-training/distillation, modern world models, artificial curiosity, and LSTM co-author — gave the keynote and a fireside chat with CEO David Ha, who argued that foundation models are among the fastest-depreciating financial assets in history and that scaling alone isn't the only path: "intelligence is born from constraints." Sessions also covered open-ended autonomous discovery (Robert Tjarko Lange) and world models & physical AI (Stefania Druga, Sebastian Risi); a full English report is available.
Related event: Sakana AI Announces In-house LLM Sawara, First Small Model Due Spring 2027(5 posts)→
More from Companies & People
- Every's first Thesis: 2027 conference streams free Nov 5 with OpenAI, Sequoia, Notion — every · 2026-11-02
- Kai-Fu Lee takes 'AI Native' book tour to New York with four events in November — kaifulee · 2026-11-02
- OpenAI, ADNOC and G42 Announce Energy-AI Partnership at ADIPEC 2026 — ingliguori · 2026-11-02
- Karpathy's HackerTube tops HN with 2677 points as Wikihack and clones flood the front page — Aizkmusic · 2026-11-02
- Transformer and NYMag unveil AI Power List: 26 people quietly running AI — ShakeelHashim · 2026-11-02
- A complete math roadmap for machine learning: linear algebra, calculus, probability — TivadarDanka · 2026-11-02