Extracting real signals from the AI noise
This week was packed with developments. Domestically, Baidu launched ERNIE 5.1 and DeepSeek funding news dominated headlines. Internationally, Google’s Veo 3 pushed AI video to new heights while NVIDIA’s Jim Fan declared VLA “dead.”
Baidu ERNIE 5.1: 6% Training Cost, #1 on Search Leaderboard
On May 9, Baidu officially released ERNIE 5.1. Its “multi-dimensional elastic pre-training” technology compresses total parameters to about 1/3 and active parameters to about 1/2, using only 6% of the pre-training cost of comparable models.
On LMArena, ERNIE 5.1 scored 1223 points, ranking #1 domestically and #4 globally on the search leaderboard. Agent capabilities exceed DeepSeek-V4-Pro, creative writing matches Gemini 3.1 Pro.
DeepSeek $50 Billion Funding
Reports suggest Liang Wenfeng personally invested 20 billion yuan, with V4.1 scheduled for June.
Google DeepMind’s “AI Co-Mathematician”
Oxford professor Marc Lackenby solved a decades-old group theory problem using this system. It scored 48% on FrontierMath Tier 4, surpassing GPT-5.5 Pro’s 39.6%.
NVIDIA’s Jim Fan: VLA Is Dead
At Sequoia AI Ascent 2026, Jim Fan proposed the “World Action Model” paradigm - essentially applying LLM training approaches to robotics.
Google Veo 3: AI Video with Audio
8-second 720p videos with synchronized sound effects and audio dialogue - a first for Google’s AI tools.
AI Pulse - Not an info搬运工, only valuable signals