🔥Telegram İfşa

Proud to announce Sakana AI’s partnership with Hokkoku Bank, based in the Ishikawa Prefecture of Japan. Sakana AI will deliver bank-specific AI-powered tools to Hokkoku Bank. We aim for this partnership to serve as a model case for other regional banks in Japan.

arxiv.org/abs/2505.22954

We’re excited to introduce Text-to-LoRA: a Hypernetwork that generates task-specific LLM adapters (LoRAs) based on a text description of the task. Catch our presentation at #ICML2025! Paper: arxiv.org/abs/2506.06105 Code: github.com/SakanaAI/Text-… Biological systems are capable of

Text-to-LoRA: Instant Transformer Adaption arxiv.org/abs/2506.06105 Generative models can produce text, images, video. They should also be able to generate models! Here, we trained a Hypernetwork to generate new task-specific LoRAs by simply describing the task as a text prompt.

Text to models!

Very excited to share this work. Here's an example of how to interact with T2L through a webui provided in the repo github.com/sakanaai/text-… More results in this 🧵

Introducing ALE-Bench, ALE-Agent! Towards Automating Long-Horizon Algorithm Engineering for Hard Optimization Problems Blog: sakana.ai/ale-bench/ Paper: arxiv.org/abs/2506.09050 ALE-Bench is a coding benchmark primarily focused on hard optimization (NP-hard) problems. We

Sakana AI developed a new coding agent, ALE-Agent, trained to solve NP-hard optimization problems. Our agent participated in a live coding competition, the challenging AtCoder Heuristic Contest, and ranked #21 out of 1,000 human participants! Learn more: sakana.ai/ale-bench/

AI will soon master Codeforces. So, what's the next challenge? 🚀Introducing ALE-Bench (ALgorithm Engineering Benchmark) 🏆 A new frontier benchmark for algorithmic coding, designed to test long-horizon reasoning on complex problems through trial and error. 🤖What is ALE-Bench?

Super happy and proud to share our novel scalable RNN model - the MesaNet! This work builds upon beautiful ideas of 𝗹𝗼𝗰𝗮𝗹𝗹𝘆 𝗼𝗽𝘁𝗶𝗺𝗮𝗹 𝘁𝗲𝘀𝘁-𝘁𝗶𝗺𝗲 𝘁𝗿𝗮𝗶𝗻𝗶𝗻𝗴 (TTT), and combines ideas of in-context learning, test-time training and mesa-optimization.

How do different coding agents perform file edits? 📝 Great blog by Fabian Hertwig 🧑‍💻: fabianhertwig.com/blog/coding-as… Different instruction tuning protocols imply that there is no clear winning approach. Most agents are either model-specific (Codex/Claude Code) or deploy robust

Introducing Reinforcement-Learned Teachers (RLTs): Transforming how we teach LLMs to reason with reinforcement learning (RL). Blog: sakana.ai/rlt Paper: arxiv.org/abs/2506.08388 Traditional RL focuses on “learning to solve” challenging problems with expensive LLMs and

Reinforcement Learning Teachers of Test Time Scaling In this new paper, we introduce a new way to teach LLMs how to reason by learning to teach, not solve! The core idea: A teacher model is trained via RL to generate explanations from question-answer pairs, optimized to improve

Very excited to share our latest work on Automated Design of Agentic Systems (ADAS) and Darwin Gödel Machine (DGM) with 机器之心 JIQIZHIXIN! Watch the interview: bilibili.com/video/BV1vJK8z… (in Chinese)