hardmaru

Excited to announce Sakana AI’s Series B! 🐟 From day one, Sakana AI has done things differently. Our research has always focused on developing efficient AI technology sustainably, driven by the belief that resource constraints-not limitless compute-are key to true innovation. This principle extends directly to our business model, which we built with that same focus on sustainability and profitability. Our company is dedicated to deploying our research into Japan’s key business and public sectors. I’m so proud of the fantastic, growing enterprise AI business we’ve built over the past year in Japan. We have partnered with some of the largest enterprises in Japan, focusing on AI applications that deliver real, practical benefits to our clients. This funding will accelerate our mission: to develop frontier AI sustainably and implement technology that truly benefits Japan. We are honored to receive support from new and existing investors, including MUFG, Khosla Ventures, Factorial, Macquarie Capital, Fundomo, Mouro Capital, NEA, Geodesic, Lux Capital, Ora Global, MPower Partners, Shikoku Electric, and IQT.

@graceisford @Lux_Capital @james_riney @coral_capital @anduriltech @AppliedInt @SakanaAILabs Japan is Back 🇯🇵🚀🔥

Awesome to bring together @Lux_Capital friends in Tokyo w/ @james_riney @coral_capital 🇯🇵 Discussed importance of BUILDING with Japan 🏗️ Critical supply chains partnering w/ US cos @anduriltech @AppliedInt 🐟 Sovereign AI lab @SakanaAILabs 🌟 Global ambition - the next SONY!

Building AI agents for real world banking workflows is incredibly difficult. It requires structuring the implicit knowledge of veteran bankers. We just published a behind the scenes look at how our Applied Team built the MUFG AI Lending Expert. They explain how we adapted concepts from our research on ALE Agent and The AI Scientist to handle complex enterprise workflows. Taking AI from the lab to a major bank is not just about better prompts. The team even used AI to process nearly 1,500 pieces of human feedback, creating a high speed improvement loop that allowed the system to scale and adapt rapidly. This interview is a great look at the engineering and product culture we are building at Sakana AI. If you want to see how we tackle hard engineering challenges and build systems for mission critical environments, I highly recommend giving it a read. Blog (Japanese): sakana.ai/mufg-ai-lendin…

@silviasetitech Prime real-estate 🗼

@hardmaru nice geoguessing😏

ｵﾋｨｽ綺麗すぎて草

@recatm Thanks!🙏

Judea Pearl: The Three Layer Causal Hierarchy （珀尔 1980年将贝叶斯算法引入ML，2011年图灵奖，他的因果论前两年也出中文版了） Pearl认为当前的机器学习系统主要学习的是统计知识，这带来了严重的理论限制，使得计算机几乎不能“对干预和回顾进行推理”，因此“不能作为强人工智能的基础”。为了说明这个问题，Pearl列举了机器学习无法完成的七类任务。这七类任务按照属性可以归为3层，也就是下图的hree Layer Causal Hierarchy 第 1 层处理观察到的相关性：看到 X，Y 的概率会怎样变化。 第 2 层处理干预后的结果：如果把 X 设成某个值，Y 会怎样 第 3 层处理反事实问题：已经发生了 X 和 Y，如果当时不是 X，会不会还是 Y 或 “如果当时没有那样做，会怎样”。 举个例子： Association：我发高烧了。我很可能得了流感。 Intervention：我吃了抗病毒药，然后我退烧了。 Counterfactuals：我的高烧是抗病毒药治好的吗？但如果我没吃，是不是也会退烧。可能我康复是因为我免疫系统已经起作用了，也可能是因为我根本得的就不是流感。 LLM 天然擅长从大规模观测文本里学习“看到 X 时通常会伴随什么”，文本里大多记录的是： - 某件事发生了什么？（哪些概念共现） - 某些变量经常一起出现？（哪些叙述常跟在什么上下文之后） - 人们如何描述经验中的规律？（哪些解释在语料里经常被当作合理解释） 这类学习可以产生很强的“表面因果线索” 而不是真正的因果建模。 LLM 拟合的是语言序列的条件分布, 其目标函数往往鼓励模型压缩大规模观测数据中的稳定统计模式。 而后两层都超出了纯 next-token prediction 能稳定提供的信息，需要强的结构性假设。只有能处理反事实，才接近人类在解释、归责、追问原因时的推理方式。 反事实被置于结构顶层，是因为它可以涵盖干预性问题和关联性问题。如果我们的模型有回答反事实问题得能力，那它同样能回答有关干预和观察的问题（反思性推理）：比如 如果价格是当前价值的两倍，会发生什么? 而一旦我们能回答介入性问题，联想性问题就能得到回答；忽略行动的部分，让观察取而代之。 哇，然后感觉跟我最近读的 @ylecun @hardmaru @SchmidhuberAI 的 世界模型的论文联起来啦😍😍😍 PEARL, J. (2000). Causality: Models, Reasoning, and Inference. Cambridge University Press, New York. 2nd edition, 2009.

“When AI Discovers the Next Transformer” Robert Lange (Sakana AI) joins Tim Scarfe (@MLStreetTalk) to discuss Shinka Evolve, a framework that combines LLMs with evolutionary algorithms to do open-ended program search. Full Video: youtu.be/EInEmGaMRLc

@tnkcoder 😋

激ウマ

@alltbl @HirokiSayama Congrats! Many of us from @SakanaAILabs also watched the lecture live 🚀

My last lecture in Komaba: youtube.com/live/V0kyaQMVd… The talk will be streamed live on this YouTube channel. @HirokiSayama

@artjng @acm_chi Congrats!

Happy to share that my PhD paper "Scene2Hap: Generating Scene-Wide Haptics for VR from Scene Context with Multimodal LLMs" has received a Best Paper Award (top < 1%) out of 6,730 submissions at ACM CHI (@acm_chi), the most prestigious conference in the human-computer interaction field🙌 Scene2Hap is an LLM-centered system that automatically designs object-level vibrotactile feedback for entire VR scenes based on objects' semantic attributes (e.g., whether and how the object vibrates) and physical context (e.g., the object's density, spatial relationships). It then renders real-time haptic feedback across the scene, calculating vibration propagation based on LLM-inferred material properties. To the best of our knowledge, this is the first paper to address the problem itself: "designing haptic characteristics of a whole VR scene with one click." Thanks a lot to my co-first-author @EasaAliAbbasi, Sara Safaee, @FKeeL1, and my advisor Jürgen Steimle!

Sakana AI has been selected by the Defense Innovation Technology Institute, under Japan's Ministry of Defense 🇯🇵, for a major multi-year research contract. This is a significant milestone for us. Alongside Finance, Defense and Intelligence is now a primary core focus area for our company. We are building an integrated system to analyze massive amounts of multimodal data across land, sea, and air domains. By leveraging our autonomous AI agents and Small Vision-Language Models (SVLM) designed for high-speed processing on edge devices like drones, we aim to automate the pipeline from raw data observation to information integration. The ultimate goal is to modernize Command and Control (C2) systems, enabling rapid situational awareness and optimal resource allocation in critical environments. National security relies heavily on information superiority. As a Japan-based AI company, we are proud to establish a dedicated domestic team of experts to ensure the secure implementation of our research. We are officially scaling our commitment to strengthening Japan's security infrastructure through Japan’s technological autonomy. You can read the full details in the announcement below (in Japanese):

#software-engineer-research-and-development" target="_blank" rel="nofollow noopener">sakana.ai/careers/#softw…

As AI makes coding more efficient, Jevons Paradox kicks in. The cost of building software is dropping, which means the demand for great Software Engineers to build even more ambitious systems is higher than ever. We are actively hiring more Software Engineers at Sakana AI to help us build these systems. Come join us in Tokyo 🗼🇯🇵 #software-engineer-research-and-development" target="_blank" rel="nofollow noopener">sakana.ai/careers/#softw…

In an alternate timeline we’d be using Evangelion GUI designs rather than CLIs

@okoge_kaz Congrats on the amazing progress on Swallow LLM, looking to this talk with @rioyokota! Looking forward to seeing what you will build during your upcoming internship @NVIDIAAI 🚀

Dive deep into the engineering of Swallow LLMs, Japan’s sovereign AI, built on Qwen3 and GPT‐OSS with Megatron‐LM & custom datasets! At #NVIDIAGTC, join us to explore the cutting-edge training behind Swallow LLMs' Japanese fluency. Explore our open outcomes: swallow-llm.github.io/index.en.html 📆Mar 17 | 4:00 PM PT 🔗Session: nvda.ws/3Mdl5bz

三菱UFJ銀行とサカナAI、融資稟議書をAIで作成　26年度にも実証実験 nikkei.com/article/DGXZQO…

Sakana AIと三菱UFJ銀行の「AI融資エキスパート」が次のフェーズへ進みます。The AI Scientistなどの私たちのコア技術が、単なる研究にとどまらず、エンタープライズの現場で実運用に向けて動き出しています。AIエージェントを実際の複雑な金融業務に適用する大きな一歩です。ぜひ読んでみてください。

@grouse324st @SakanaAILabs ようこそ🐡

Sakana AI @SakanaAILabs にApplied Research Engineerとして入社しました！ 生成AIなど最先端の研究成果の事業応用を頑張っていきます！！！！

AIの実装とオペレーションの乖離。AI開発会社サカナAIのデビット・ハさんが業務をタスクとジョブに分解する必要性を述べています。前者をAIで代替、後者に人間を専念させる。生産性や付加価値を上げるってこういうことなんですね。実務でも気づきがありました。時間をみつけてnoteにまとめてみます😊。 #AI #生産性 nikkei.com/article/DGXZQO…

I recently had the pleasure of joining @nikkei podcast to discuss Sakana AI’s vision, the current landscape of AI development, and the strategic path forward for Japan 🇯🇵 As the AI industry moves incredibly fast, I believe it’s crucial to step back and think about how we build these systems responsibly and effectively. During the conversation, we touched on a few key themes that are central to what we do at Sakana AI: 🇯🇵 Japan’s “Hybrid Strategy” The optimal path forward isn’t about relying 100% on foreign-made AI, nor is it about closing ourselves off with purely domestic models. The key is a hybrid approach—striking the right balance and combining the best technologies created both at home and abroad. 🏯 Why “Sovereign AI” Matters Building domestic AI capabilities goes far beyond national security. I strongly believe that every country should aim to develop its own AI technology to protect, preserve, and pass down its unique cultural identity, languages, and traditions. 🌱 Towards Sustainable AI Development We are still at a stage where we must ask ourselves: Will the massive capital currently being poured into AI truly pay off in the long run from a sustainability standpoint? At Sakana AI, we believe that enormous compute resources aren’t always required to train a new foundation model. Efficiency matters, and there is still immense room for innovation everywhere. You can read the full article and listen to the podcast episode (in Japanese) here: