Introducing ShinkaEvolve, an open-source approach to sample-efficient LLM-driven program evolution 🧬 Paper: https://t.co/Q1PhLaZZDV Blog: https://t.co/MpVNwyNFZv Code: https://t.co/pZMzQOeoaX The AI Scientist, Darwin Goedel Machine, and AlphaEvolve have fundamentally shaped

Kicking off #TEDAISF2025 with none other than @YesThisIsLion — Co-Author of “Attention Is All You Need” and CTO/Co-Founder of Sakana AI. His reflections on the evolution of AI architecture and what’s next for intelligence, made for the perfect Session 1 opener. 🚀

Saturday’s plans = reptiles. See you at Reptiday Sarasota!

Sakana AI has just leveraged their evolutionary code optimization system, ShinkaEvolve, to earn the 1st prize at @icfpcontest2025 🏆 ShinkaEvolve enabled up to a 10x speedup by evolving clever SAT encodings, unlocking solutions to far larger and more complex problems than

👀 for all LLM-driven scientific discovery enthusiasts -- there might be some content to watch out for 🙈 Thank you, @ecsquendor and Abby for having me on 🤗

Stoked to see ShinkaEvolve support @SakanaAILabs 's @iwiwi's Unagi team win the 2025 ICFP Programming Contest 🎉 ShinkaEvolve is an LLM-driven evolutionary program optimization system that discovered efficient auxiliary variables for a downstream SAT solver 🚀 Code:

A little secret behind our recent win 🥇 at the ICFP Programming Contest 2025: we leveraged ShinkaEvolve, an evolutionary AI system we're developing at Sakana AI! 🧬 It didn’t just auto-boost our score; we actually learned from its ideas. Full story:

Competitive programmers collaborated with Sakana AI’s ShinkaEvolve to win 1st place in the 2025 ICFP Programming Contest 👏 https://t.co/okBe1uI2Zl What happens when competitive programming experts master a cutting-edge AI tool? Team Unagi, which includes Sakana AI's Research

Proud to share that my team, Unagi, won the ICFP Programming Contest 2025! 🏆 Pushing SAT solvers to their limits was a lot of fun. This is our 8th win. Grateful to my teammates: @imos @sulume @wata_orz @toslunar @chokudai https://t.co/qjqE7D414g https://t.co/pI9KdqCb7Y

I had a lot of fun participating in one of the last World Expo panels during the previous weekend in Osaka. I talked about our work at @SakanaAILabs on automating scientific discovery, open-endedness, and collective debugging ;) Thank you, @nrryuya, @RishiBommasani and

Super fun to watch our very own AI Scientist-v2 knocking off one of the 2024 predictions for AI Scientists!

Super cool work on JAX-accelerated Arcade Games 🕹️ using Chip-8 ROM emulation by @riiswax et al. ⚡ More than 20 games allowing for end-to-end parallelization with up to 14x speedups vs. CPU-based alternatives. Paper: https://t.co/InW9LNy3KP Code: https://t.co/c5w21oIKCb

【お知らせ】 大阪・関西万博のプログラム「人とAIの共生：新たな知能とどう向き合うのか」にLayerX AI・LLM事業部長 の中村龍矢が登壇 します。 https://t.co/d0wt98ccr5

SakanaAIがShinkaEvolveを発表。ShinkaでEvolveだったら進化進化やん。これは以前のAIサイエンティストの発展版で、以前は総当たり的な探索してたのを進化的なアルゴリズムを組み合わせてかなり探索効率が上がったらしい

Something that is truly mesmerizing about LLM-driven evolutionary optimization is the accumulation and composition of stepping-stone programs. Instead of greedily hillclimbing the best program, Shinka explores diverting mutations and discoveries diffuse over the tree search 👨‍🔬

【Sakana AIがオープンソース進化フレームワーク「ShinkaEvolve」を公開】 Sakana

ShinkaEvolve出ました！LLMを使ったコード自動改善のフレームワークです。Sakana AI版AlphaEvolve……と言うと話は簡単ですが、性能も使いやすさも様々な工夫があり、凄く良く出来てます。自分も面白い利用を既に何度か試してまして、このソフトウェアの大ファンです。その話はまた後日……！

We release ShinkaEvolve under the Apache 2.0 license and are excited to see what you explore! Here is a notebook to get started with:

To make usability as smooth as possible, ShinkaEvolve comes with a simple-to-use WebUI allowing you to monitor discovery runs in real-time! Simply run ` shinka_visualize --open`

Improving Competitive Programming Solutions on ALE-Bench We investigate whether Shinka can work in tandem with solutions discovered by other agents. More specifically, we use AtCoder heuristic programming solutions discovered by Sakana’s ALE-Agent and search for improvements

We evaluated ShinkaEvolve's applicability on vastly different additional problems: Automated Design of Agent Scaffolds for math reasoning on AIME The Shinka-evolved agent implements a three-stage architecture leveraging diverse expert personas, critical peer review, and

With ShinkaEvolve, we aim to make a big step towards improving efficiency and broad accessibility to automated computational discovery. It combines multiple key algorithmic improvements: 1) An adaptive parent program sampling strategy balancing exploration & exploitation. 2) A