David Bennett retweeted
AI推論の高速化には、ハードウェアの高速化だけでは足りません。
@aiand_ は、@nvidia・@AMD・@tenstorrent向けにMiniMax H3を最適化しました。
→ キャッシュ：リファインメント1ステップの処理時間を2.53秒から0.05秒に短縮
→ ステップ蒸留：15秒動画の生成時間を229秒から19秒に短縮
適切なハードウェアと最適化手法の組み合わせによって、より高速かつ低コストな動画生成を実現しました。
🇯🇵 技術ブログはこちら：aiand.com/jp/newsroom/road-t…
David Bennett retweeted
本日、@aiand_ videoの提供を開始しました。高速かつコスト効率の高いAI動画生成が、ai& inferenceに加わりました。
テキスト、写真、既存の動画から、新たな映像を最短12秒で生成。従量課金制で、H3 Fastは生成動画1秒あたり$0.0125から。
@nvidia・@AMD・@tenstorrent向けに最適化されたMiniMax H3が、日本国内のai&インフラ上で稼働します。
こちらからお試しください：aiand.com/jp/video
発表の詳細はこちら：aiand.com/jp/newsroom/aiandv…
David Bennett retweeted
Today we launched @aiand_ video, bringing fast, cost-efficient video generation to ai& inference
Turn text, photos, and existing clips into new footage in as little as 12 seconds, with pay-as-you-go pricing starting at $0.0125 per generated second on H3 Fast.
@MiniMax_AI H3 optimized across @nvidia, @AMD, and @tenstorrent, running on ai& infrastructure in Japan.
Try it at aiand.com/video
Read the full announcement here 👉aiand.com/newsroom/aiandvide…
David Bennett retweeted
Applications are open for @aiand_ x @AMD Build Day! 🗼
8 problems from AMD partners. 8 teams. 4.5 hours to build. Then demo your work on the Advancing AI stage.
Build with @Kimi_Moonshot, @Zai_org GLM, @Alibaba_Qwen, and @MiniMax_AI on ai& inference, powered by AMD.
📅 December 1, 2026
🕤 9:30 a.m. – 7:30 p.m.
📍 Palace Tokyo, Marunouchi
Apply by November 9: amd-advancing-ai.jp/buildday…
David Bennett retweeted
The guy who gave us QLoRA, says DeepSeek's inference system is on par with OpenAI's stack and better than what Anthropic runs, at least at their scale.
Now check the license. Apache 2.0.
Until today a production stack at that level existed inside a handful of companies. It was the thing that made their API fast and their margins fat, and none of them ever showed it to anyone.
Today any startup, any university, any cloud in any country can download it, read it and run it. It runs on NVIDIA Blackwell and on Huawei Ascend, so it isnt tied to 1 chip vendor either.
This is what I mean when I say open weights are about more than downloading a chatbot. DeepSeek is giving away the knowledge that only a few labs had, and that changes who can build an AI company at all.
Respect @deepseek_ai
David Bennett retweeted
ai& が明日開催される Plug and Play Japan Summit 2026に参加します。
AI Center of Excellence Startup Pitchに@mario_meissnerが登壇。@aiand_が日本国内で進める、コスト効率の高いAIインフラの構築についてご紹介します。
📅 10月21日（水）| 🕓 17:40〜18:10
📍 1F Play Stage
🔗 お申し込みはこちら：contents-jp.plugandplaytechc…
David Bennett retweeted
Congrats to @deepseek_ai on the launch of DeepInfer! 🎉
Great to see the engineering behind DeepSeek's production inference stack being open-sourced, and thanks for the shoutout to the SGLang community 🧡
We're especially excited about the work on large-batch and multi-GPU inference. Can't wait to dig deeper and bring some of these ideas into SGLang!
🚀 Introducing DeepInfer, our in-house inference framework, now fully open source!
Powering DeepSeek-V4.1-Flash at scale with high throughput and low latency. We’re sharing our production stack to help the community lower compute costs, reduce energy use, and make AI more affordable.
🔗 github.com/deepseek-ai/DeepI…
1/4
David Bennett retweeted
This is for all the writers ( and want to be writers) in this beautiful community 💖✍🏼👩💻👨🏻💻📝. This gave me some serious guffaws and I hope it does for you as well 👍🏼💙. Happy writing ✍🏼🙏🏼. @LouDPhillips @DzintraSullivan @ChrisBohjalian @therealSIR3 @hwinkler4real @AuthorFleur @donnacox1021 @drdoomed01 @jameshealyjr
David Bennett retweeted
See you at @plugandplayjp Summit 2026 this week!
Catch @mario_meissner at the AI Center of Excellence Startup Pitch to hear how @aiand_ is building cost-efficient AI infrastructure in Japan.
📅October 21 | 🕓17:40-18:10 JST
📍1F Play Stage
🔗Register here: contents-jp.plugandplaytechc…
David Bennett retweeted
@aiand_ runs open models like Kimi and DeepSeek on GPUs in Japan, behind an OpenAI and Claude compatible API. switching your agent over is one line
they're sponsoring the Tokyo Agent Hackathon on 17 Oct, so nobody's rationing credits
see you on 17th Oct :)
David Bennett retweeted
@Plus81builders Tokyo Agent Hackathonを支援できることを、大変うれしく思います！🚀
@aiand_ runs open models like Kimi and DeepSeek on GPUs in Japan, behind an OpenAI and Claude compatible API. switching your agent over is one line
they're sponsoring the Tokyo Agent Hackathon on 17 Oct, so nobody's rationing credits
see you on 17th Oct :)
David Bennett retweeted
Proud to support the @Plus81builders Tokyo Agent Hackathon tomorrow! 🚀
@aiand_ runs open models like Kimi and DeepSeek on GPUs in Japan, behind an OpenAI and Claude compatible API. switching your agent over is one line
they're sponsoring the Tokyo Agent Hackathon on 17 Oct, so nobody's rationing credits
see you on 17th Oct :)
David Bennett retweeted
デモ審査のあと、優勝チームはAMDのレセプションでパートナー企業100社以上の前で発表できるとのこと！個人参加で、テーマ別で8チームに分類されるみたい。参加してみよかなぁ。
This quoted post is unavailable.
David Bennett retweeted
日経 GenAI Summit 2026（東京）での充実した数日間となりました。🗼
@aiand_ のCEO @DavidBennett__ が、日本におけるAIインフラの構築とスケールについての議論に加わり、 @aiand_ が日本最大の推論専業プロバイダーへと成長してきた歩みをご紹介しました。
本イベントを主催された日本経済新聞社の皆様に、心より感謝申し上げます。
David Bennett retweeted
Finding a job is a job. I spent months treating mine like one. Spreadsheets, 40 tabs, follow ups I forgot to send. So I did what engineers do with messy processes. I built an app.
David Bennett retweeted
A great few days in Tokyo at Nikkei’s GenAI Summit 2026. 🗼
Our CEO @DavidBennett__ joined the conversation on building and scaling AI infrastructure in Japan, sharing how @aiand_ has grown into Japan’s largest dedicated inference provider.
Thank you to Nikkei for hosting. 🚀
David Bennett retweeted
This is big!
Super fired up that @tenstorrent and @aiand_ will support their mission with lots of compute!
On August 31, the first two participants were injected with @PopVaxIndia’s first fridge-stable RNA vaccine candidate in a Phase I clinical trial in Australia. We built this vaccine at our RNA Foundry in Hyderabad – from computational design to GMP manufacturing. Here's how:
David Bennett retweeted
RL environments are becoming a serious data discipline.
TermGrade is a great example of what that looks like:
→ 1,004 executable environments
→ 36k trajectories, including failures
→ 6 models graded on every task
→ difficulty measured for the model you're actually training
And everything is open.
This is exactly the kind of work we need more of for agentic RL.
We're releasing TermGrade at @aiand_: high-quality RL environments for terminal agents, fully open source, along with the full story of how we built and graded them, and the RL recipe we trained with.
That's 1k executable environments, each verified by execution and graded against 6 models, plus all 36k trajectories behind the grades, failures included.
And training gemma-4-31B-it on the tasks it solved half the time took it +3.1 on Terminal-Bench 2.1.
Thread below 🧵