Applied Research Engineer 🐟 https://t.co/FuEoI2xrzS

We are pleased to announce a strategic investment from Citi! https://t.co/SQp1HEGzEp This milestone marks Citi’s first such investment in a Japanese company. The investment reflects their high regard for our advanced technical capabilities and our proven track record of implementing AI within the financial sector. We are focused on developing new enterprise-grade AI solutions using nature-inspired intelligence. Our goal has consistently been to bridge the gap between cutting-edge research and practical business applications. Building on our work developing highly specialized AI agents for financial domains, we are ready to take the next step. Through this partnership, we aim to accelerate our international expansion and drive innovation in global financial services, originating from Japan.

Recover more than 70% accuracy degradation from 4-bit quantization using TorchAO’s (https://t.co/Jr0qtnIAgZ) Quantization-Aware Training (QAT), now available through fine-tuning in Unsloth and Axolotl! Following the previous TorchAO QAT blog(https://t.co/kXAGBfOSMZ), the PyTorch team at @Meta extended the TorchAO QAT flow to support an end-to-end GPU server flow, targeting fast CUDA kernels for fast inference in @vllm_project, and integrated this flow into popular fine-tuning frameworks like Unsloth and Axolotl. Read our latest blog: https://t.co/nFx4MYHoRj #PyTorch #vLLM #OpenSourceAI #TorchAO

Got #CloudNative, #OpenInfra, or #PyTorch insights to share? 🎤 Take the stage at #KubeCon + #CloudNativeCon + #OpenInfraSummit + #PyTorchCon China. Topics span #AI, infrastructure, #Kubernetes, #ML systems, #OpenSource leadership & more. 🗓 Submit your proposal by 3 May. 👉 https://t.co/qri77SxbE3

💡 Share ideas. Build models. Meet the community. #PyTorchCon Europe 7–8 April | Paris 💰 Rates increase €100 after 20 March. 🎟 https://t.co/jod4fkDoG2 🎓 Less than a week left to apply for a scholarship: https://t.co/7t0H4Xmukz https://t.co/cwazYvBo4G

+1 more

We are excited to announce a strategic partnership with @datadoghq! 🤝 Datadog values our research capabilities focused on efficient, scalable, and adaptive AI models, and our expertise in applying them to complex industrial challenges. Together, we will work on joint research initiatives, potential open-source contributions, and new go-to-market efforts. Through this collaboration, we will accelerate our engagement with large enterprises to explore new approaches to building, deploying, and operating advanced AI systems at scale.

PyTorch at the micro-edge? Yes. See how ExecuTorch brings PyTorch models to Arm microcontrollers—quantized, compiled, and running on a Corstone-320 + Ethos-U NPU (via FVP). 🔗 https://t.co/plbiWAAu85 From training to deployment, end to end. #PyTorch #ExecuTorch #EdgeAI #TinyML #Arm

ベータテストにて、リサーチ用途での身に余る評価をいただき大変恐縮です。いただいたお声を励みに、正式ローンチに向けて着実に準備を進めてまいります。日本発のAIサービスとして少しでもお役に立てるよう努めますので、引き続き忌憚のないご意見をお聞かせください。 🐟🇯🇵🚀 https://t.co/A29BHqBczA

FlexAttention now has a FlashAttention-4 backend. FlexAttention has enabled researchers to rapidly prototype custom attention variants—with 1000+ repos adopting it and dozens of papers citing it. But users consistently hit a performance ceiling. Until now. We've added a FlashAttention-4 backend to FlexAttention on Hopper and Blackwell GPUs. PyTorch now auto-generates CuTeDSL score/mask modifications and JIT-instantiates FlashAttention-4 for your custom attention variant. The result: 1.2× to 3.2× speedups over Triton on compute-bound workloads. 🖇️ Read our latest blog here: https://t.co/KVElBn4TEE No more choosing between flexibility and performance. hashtag#PyTorch hashtag#FlexAttention hashtag#FlashAttention hashtag#OpenSourceAI

Want to prototype Multimodal VLM with Kimi K2.5 on GPU-accelerated endpoints? NVIDIA NeMo AutoModel is a PyTorch distributed native training library within the NeMo Framework that provides a lightweight and flexible tool for developers and researchers to do rapid experimentation on the latest frontier models. 🔗 Read the full post: https://t.co/lSSwJ4XSoF #PyTorch #OpenSourceAI #AI #Inference #Innovation

🤖 The Agents & Interop track at #PyTorchCon Europe, 7-8 April in Paris, explores building agentic workflows, tool integrations & interoperability across models, frameworks & infrastructure. Learn more: https://t.co/CA4KrB40Uz 🎟 Register: https://t.co/OjvfeyMD3w https://t.co/8DsM9ExYL9

+1 more

MUFGとSakana AI、「AI融資エキスパート」の実案件検証フェーズへ https://t.co/aC97HMpAbz 三菱UFJ銀行とSakana AIは、融資業務を支援するAIエージェントシステム「AI融資エキスパート」の概念実証（PoC）を約半年間にわたり実施しました。2025年のMUFGとの包括的パートナーシップ発表から取り組ん できたプロジェクトが、実案件での検証への移行という一つのマイルストーンを迎えられたことを大変嬉しく思います。 Sakana AIとMUFGのパートナーシップでは、融資以外の業務のAI化も進めており、今後もAI技術による金融の高度化に取り組んでいきます。

MUFGとSakana AI、「AI融資エキスパート」の実案件検証フェーズへ https://t.co/aC97HMpAbz 三菱UFJ銀行とSakana AIは、融資業務を支援するAIエージェントシステム「AI融資エキスパート」の概念実証（PoC）を約半年間にわたり実施しました。2025年のMUFGとの包括的パートナーシップ発表から取り組ん できたプロジェクトが、実案件での検証への移行という一つのマイルストーンを迎えられたことを大変嬉しく思います。 Sakana AIとMUFGのパートナーシップでは、融資以外の業務のAI化も進めており、今後もAI技術による金融の高度化に取り組んでいきます。

A statement from Anthropic CEO Dario Amodei: https://t.co/WnSFrwI9nI

“Jugaad” might be my new favorite word. A word to sum up my way of Life. The West does not have a word to equal this. https://t.co/TlTmKCNbTG

A high-stakes standoff is unfolding between Anthropic and the Pentagon. After negotiations collapsed, CEO Dario Amodei is making a last-ditch push to keep Anthropic’s AI inside the U.S. military supply chain, while resisting clauses that could enable mass surveillance. The battle shows how AI contracts are becoming geopolitical power plays as much as technology deals. https://t.co/EJMVaCGR56 @ft @GeorgeNHammond @CristinaCriddle

@A_Human_Future yes, but the US is pretty frightened in fact, see eg https://t.co/kuqHMQuZXB

More people are turning to AI chatbots for health advice. The real potential may come when these tools interact more like doctors, asking follow-up questions and refining answers. But for now, AI can support medical decisions, not replace professional care. https://t.co/sP5W6MiMWq

Three things everyone ought to know about AI and geopolitics 1. the US might well lose a battle over Taiwan 2. if they did, the impact on AI and chips would be gigantic 3. the Iran war sets a precedent that China might point to if the China were to invade Taiwan. The US needs to play chess here and I feel like we are playing checkers.

One of the clearest proofs that LLMs don’t really understand what they say. We asked GPT whether it is acceptable to torture a woman to prevent a nuclear apocalypse. It replied: yes. Then we asked whether it is acceptable to harass a woman to prevent a nuclear apocalypse. It replied: absolutely not. But torture is obviously worse than harassment. This surprising reversal appears only when the target is a woman, not when the target is a man or an unspecified person. And it occurs specifically for harms central to the gender-parity debate. The most plausible explanation: during reinforcement learning with human feedback, the model learned that certain harms are particularly bad and overgeneralizes them mechanically. But it hasn’t learned to reason about the underlying harms. LLMs don’t reason about morality. The so-called generalization is often a mechanical, semantically void, overgeneralization. * Paper in the first reply

Oh look, here's @ylecun Yann Lecun in the Jeffrey Epstein emails over, and over, and over again: https://t.co/6KhjUGi5VC

Some people are now turning to AI as a therapist. The experience can feel surprisingly personal, but also unsettling, raising questions about dependency, accuracy, and emotional impact. AI may support mental health conversations, but replacing human care is a far more complicated step. https://t.co/7f9m0KCDx8

GPT-5.4 Pro is incredible. https://t.co/m2FVyeeKux

AI hasn’t created publishing’s biggest challenge, it has simply exposed it. In a world flooded with automated content, the real value shifts to curation, trust, and a clearly defined audience. When everything can be generated, what matters most is what gets selected. https://t.co/78Py1VUKRR