Ethan He

Excited to announce: I'm joining @xai to help accelerate humanity's quest to understand the universe! With @grok 4 unlocking new frontiers in AI, I can't wait to dive in and push the boundaries of what's possible. 🚀

At omakase today, the chef said mastering the art of sushi takes decades—even rice demands seasonal tweaks in water temp & vinegar balance. 🍣 Training LLMs is the same: as data shifts, so must your hyper‑params. Copy‑pasting settings from papers doesn't work. Craft, iterate,…

What do sushi and AI inference have in common? The right temperature. 🍣🤖 I got inspired by today’s sushi (yummy!) AI temp = randomness: ▪️0.2 = deterministic and reliable; for math and summaries ▪️0.5 = balanced; for marketing and user-friendly responses ▪️0.9 = creative,…

After 2 years at @nvidia, I’m writing to share that I’ll start a new adventure. Working with brilliant teammates on cutting‑edge AI has shaped me so much: - Cosmos debuted as a SOTA world model and earned 8 k⭐️ on GitHub. - We open‑sourced the first recipe for upcycling 100 B+…

📢 New blog post: Reinforcement Learning with NVIDIA NeMo‑RL – Reproducing the DeepScaleR Recipe with GRPO 🚀 We just published a deep‑dive showing how NeMo‑RL, our new open‑source post‑training library in the NVIDIA NeMo framework, scalably trains reasoning models from a…

🚨 New open‑weight LLM drop → T5Gemma 🚨 Google just published a collection of encoder‑decoder models that were adapted from their decoder‑only Gemma‑2 checkpoints instead of being trained from scratch. The idea is simple but provocative: copy the pretrained weights into an…

Grok-4 release is mind-blowing! A few things that strike me: •Test-time compute scaling with multi-agent interactions, where agents actively collaborate and challenge each other. •Advanced tool use during training, with @elonmusk’s ambition to integrate even more complex tools…

🚀 Need to serve millions of tokens in real time? Meet Helix Parallelism. Problem: Long‑context LLMs choke on two bottlenecks—📚 KV‑cache streaming that hammers DRAM bandwidth, and 🏋️ FFN weight reads that stall every token. Solution: Helix Parallelism, co‑designed for the…

Submit your proposals w/ the #NVIDIAAcademicGrant Program! 🤖 Agentic Model Systems & Robotics 💡 Generative AI Alignment & Inference 🔬 Foundation Models for Chemistry & Climate Science 📅 Apply by September 30: nvda.ws/4l4iM6E bit.ly/3IcH7sr

NeMo AutoModel is an NVIDIA-developed library that delivers a high-performance, easy-to-use solution for fine-tuning and pretraining large language models (LLMs) and vision-language models (VLMs) directly from the Hugging Face Hub. It provides true Day-0 compatibility with any…

NEW! DGX Cloud Benchmarking Recipes for NVIDIA GB200—optimize AI workloads, cut costs, and maximize performance. Try now! tinyurl.com/dmyvkh3n bit.ly/4nm5Uua

World's first autonomous delivery of a car! This Tesla drove itself from Gigafactory Texas to its new owner's home ~30min away — crossing parking lots, highways & the city to reach its new owner

Gemma 3n models now run on NVIDIA Jetson & RTX systems with powerful new audio capabilities—enabling richer multimodal experiences. Jetson devs using Gemma 3n for social good can join DeepMind’s Gemma 3n Impact Challenge on Kaggle. bit.ly/3HZoFn7

Join an in-depth conversation with NVIDIA WIT members as they share their pivotal moments and how they took risks, navigated career shifts, and empowered themselves—and each other. 🗓️ June 26 | 🕓 5–6 p.m. PT 🔗 Register: nvda.ws/3TpeeLV bit.ly/3Tvr8s5

A year ago, I highlighted @Tesla surpassing 1 billion miles driven using Full Self-Driving (FSD). At the time, many expressed skepticism in the comments. Today, Tesla launched its @robotaxi service. While it currently operates only within a limited geofenced area in Texas, it…

Tesla Robotaxi: A New Era Begins I’ve (very fortunately) been part of multiple robotaxi launches. But this one is different and feels much more profound. It’s a paradigm shift. It’s the GPT moment for real-world autonomy. Tesla’s robotaxi runs vision-only -- no lidar, no radar,…

Just listened to the first @OpenAI podcast featuring @sama. Here are some points I found interesting: Privacy and Trust: Users currently trust ChatGPT significantly more than traditional search engines like @Google. However, this trust could diminish if OpenAI prioritizes…

GPT4 killed @heyjasperai. GPT record mode is a game changer to transcription apps like @otter_ai ☠️