OpenAI stumbles, China surges & more (April 28, 2025)

Good morning!

From China's AI sprint to Meta's Llama heading to space, this week’s AI news was packed with big moves, surprising tech drops, and a few space oddities.

Let’s dive in.

🧠 Major Headlines

OpenAI’s o3 Model Falls Short

• OpenAI’s o3 model underperforms compared to the earlier o3-preview, especially in reasoning benchmarks (ARC-AGI-1)
• Sam Altman promises fixes for GPT-4o's "sycophant-y" tone, targeting more natural conversation

China Accelerates AI Independence

• Xi Jinping announced a full national strategy for homegrown AI chips, software, and talent
• Huawei rumored to be testing NVIDIA alternatives; DeepSeek R2 expected to cut costs and rival top models

Microsoft’s Recall Tool Finally Lands

• Recall gives AI PCs a "photographic memory," letting users retrieve past documents, websites, and screenshots using simple search prompts
• It marks a major shift toward memory-driven computing experiences

Meta’s Llama Goes to Space

• Meta deployed a custom version of Llama 3.2 on the ISS to assist astronauts, highlighting the future of local AI in low-connectivity environments

⚡ Quick Hits on China’s AI

• Baidu slashes costs with new ERNIE AI models, openly targeting DeepSeek and GPT competitors
• Butterfly Effect’s Manus AI agent lands $75M in funding after outperforming big players
• DeepSeek-R2 launched — a multilingual, efficient model positioning China even closer to global AI leaders

🛠️ Tool Stack of the Week

• Tavus Hummingbird-0: Next-gen lip-syncing model
• Rime Arcana: Captures real human speech nuances with wild realism
• You.com ARI: Deep research agent that analyzes 400 sources instantly
• Sana AI: No-code agents to automate any workflow

💬 Prompt of the Day

Boost remote team productivity
"Act as a remote team productivity expert. Suggest proven hacks, async workflows, and creative tips for keeping remote teams engaged and thriving."

🎯 TL;DR

AI isn’t slowing down — from Chinese chip breakthroughs and Microsoft’s memory PCs to astronauts getting LLM copilots, the race to the future is in full throttle. Whether it’s new tools, cost wars, or LLMs in orbit, the landscape is evolving faster than ever.

That’s a wrap!

Thanks for reading — see you in the future 🌟

(And remember: next time you lose a file, maybe your computer will just… Recall it for you.)

Cheers,

David