MSFT Applied Scientist ,Hulu Applied Scientist,dedicated to learning and disseminating knowledge about AI, software engineering, and engineering management
🍌 nano banana pro prompt (with gemini)
Dynamically generate a current weather card based on a given city name.
--- prompt ---
Present a clear, 45° top-down view of a vertical (9:16) isometric miniature 3D cartoon scene, highlighting iconic landmarks centered in the composition to showcase precise and delicate modeling.
The scene features soft, refined textures with realistic PBR materials and gentle, lifelike lighting and shadow effects. Weather elements are creatively integrated into the urban architecture, establishing a dynamic interaction between the city's landscape and atmospheric conditions, creating an immersive weather ambiance.
Use a clean, unified composition with minimalistic aesthetics and a soft, solid-colored background that highlights the main content. The overall visual style is fresh and soothing.
Display a prominent weather icon at the top-center, with the date (x-small text) and temperature range (medium text) beneath it. The city name (large text) is positioned directly above the weather icon. The weather information has no background and can subtly overlap with the buildings.
The text should match the input city's native language.
Please retrieve current weather conditions for the specified city before rendering.
City name:【上海】
Fundador de Kimi.
La arquitectura exacta detrás de una empresa valorada en 20.000 millones de dólares. Probablemente no haya una forma más rápida de aprender a crear agentes de IA ahora mismo
Sin relleno. Explicado por la persona que lo construyó.
Very interesting work! Forced "wait" token may be a bit tricky to train the llm model,however it work! Niklas absolutely knows how to train the llm effeciently. That reminds me of the hard negavite samples in recommendation model study.
Last week we released s1 - our simple recipe for sample-efficient reasoning & test-time scaling.
We’re releasing 𝐬𝟏.𝟏 trained on the 𝐬𝐚𝐦𝐞 𝟏𝐊 𝐪𝐮𝐞𝐬𝐭𝐢𝐨𝐧𝐬 but performing much better by using r1 instead of Gemini traces. 60% on AIME25 I.
Details in 🧵1/9
Last week we released s1 - our simple recipe for sample-efficient reasoning & test-time scaling.
We’re releasing 𝐬𝟏.𝟏 trained on the 𝐬𝐚𝐦𝐞 𝟏𝐊 𝐪𝐮𝐞𝐬𝐭𝐢𝐨𝐧𝐬 but performing much better by using r1 instead of Gemini traces. 60% on AIME25 I.
Details in 🧵1/9
🧠 Mapping the AI Research Frontier 🚀
Created a visualization showing how modern AI systems evolve from foundation models → specialized applications!
Two key pathways :
- Full parameter tuning/LoRA 🔄
- Knowledge distillation → RAG systems 📚
Thoughts on this architecture? 👇