Sakana AI unveils pretraining from scratch with Sakana Sawara LLM series, first model due spring 2027
tkasasagi · x · 2026-11-03
Sakana AI hosted the "From World Models to Real-World AI" symposium at Hitotsubashi Hall in Tokyo on Oct 26, drawing 400+ attendees.
- Chief scientist Takuya Akiba announced Sakana Sawara (provisional name), an LLM series pretrained entirely in-house; the first small model, Sawara Small, is slated for release in spring 2027.
- Chief Scientific Advisor Jürgen Schmidhuber (joined September) keynoted on the lineage from Amari and Fukushima to his 1990 world models and artificial curiosity work, then conversed with CEO David Ha, his co-author on the 2018 "World Models" paper.
- Schmidhuber argued generative AI realized only about half of his vision; self-experimenting AI scientists and real-world Physical AI haven't scaled yet — an opening he sees for players outside the big labs, with Japan as a promising base.
More from Companies & People
- Gap runs loss-prevention AI entirely on Mac minis, cited on Apple's Q4 earnings call — akumar · 2026-11-03
- Founder playbook for 2027: your moat is proprietary AI knowledge, marketing is what's left for you — alexmacgregor__ · 2026-11-03
- Kaggle veteran joins Sakana AI full-time after 18 months as part-time contractor — SakanaAILabs · 2026-11-03
- Palantir CTO cites Anthropic's S-1: AIG's AI underwriting runs entirely on AIP — eliano · 2026-11-03
- Replit CEO Amjad Masad: next cycle's Anthropic won't look like Anthropic, VCs should back full-stack paradigm over nth LLM lab — beffjezos · 2026-11-03
- AI bubble burst favors pure ML backgrounds: end-to-end skills now beat niche expertise — kmeanskaran · 2026-11-03