Good morning, AI enthusiasts. China just unveiled SpikingBrain-1.0, proving that biomimetic AI design can dramatically outperform traditional approaches while using a fraction of the resources.
Are we witnessing the beginning of a post-transformer era in artificial intelligence?
In today's recap:
China's brain-inspired AI uses 98% less training data
Research reveals why AI gives inconsistent responses
Microsoft tests Anthropic Claude for Office 365 apps
8 new AI tools and more
CHINA & AI
China unveils brain-inspired AI
Source: VCG
Recaply: China just unveiled SpikingBrain-1.0, a new AI system that works like the human brain. Unlike energy-hungry systems like ChatGPT, this brain-inspired AI needs only 2% of the training data that normal AI systems use. Chinese scientists built it using only their own computer chips, which is a big step toward not needing foreign technology.
Key details:
SpikingBrain-1.0 runs 26 times faster than current AI systems when handling long documents.
The system uses "spiking neurons" that work like real brain cells.
SpikingBrain-1.0 is built entirely on Chinese-made MetaX C550 computer chips.
It comes in 7-billion and 76-billion parameter versions that anyone can use.
It's great for reading legal documents, medical records, and scientific papers.
Why it matters: This breakthrough could change how AI systems are built worldwide. By copying how the brain naturally works, SpikingBrain-1.0 shows that better design is more important than more data. China's success with its own technology also means more global competition in AI.
THINKING MACHINES LAB
New research explains AI's inconsistent behavior
Source: Thinking Machines Lab
Recaply: A team from Thinking Machines Lab just published research on why AI gives different answers to the same question. They found the real reason behind this problem.
Key details:
The old idea that random GPU processing causes this problem isn't the full story.
When servers are busy, they process information in batches, creating different paths for calculations.
Three key parts of AI systems need to be redesigned to fix this problem.
Horace He wrote this research for Mira Murati's new AI company.
The lab will share more findings on their blog.
Why it matters: Making AI give the same answer each time could change how people use these systems for important work. Companies need AI they can trust, and this research shows how to make that happen.
MICROSOFT & ANTHROPIC
Microsoft considers Anthropic AI for Office 365
Source: Microsoft
Recaply: Microsoft is changing its AI approach by considering Anthropic's Claude models for Office 365 apps. Tests showed Claude Sonnet 4 works better than OpenAI's models for making PowerPoint slides and Excel calculations. This means Office users will get better AI features while Microsoft doesn't have to rely on just one AI partner.
Key details:
Claude Sonnet 4 did better than OpenAI's GPT-5 at creating presentations and working with financial spreadsheets.
Microsoft will pay Amazon to use Anthropic's AI models through cloud services.
Office Copilot still costs $30 per user each month even with the new AI technology.
This partnership helps Microsoft use more than just OpenAI while keeping their current relationship with OpenAI.
Users will see better AI features in Word, Excel, Outlook, and PowerPoint soon.
Why it matters: This partnership shows a big change in how tech companies use AI. Instead of using just one AI provider, companies are now mixing different AI systems to get the best results for specific tasks.
NEWS
Beyond the big stories in AI today
Vodafone tested virtual people in TikTok ads, showing a digital woman talking about internet deals and €120 cash back offers.
STIM created the first AI music license, letting companies use copyrighted songs for AI training while paying more than 100,000 songwriters.
Mozilla added a feature to Firefox on iPhones where you can shake your phone to get AI summaries of web pages (if they're under 5,000 words).
OpenAI updated ChatGPT's developer mode to support Model Context Protocol, making it easier to connect with external tools.
Modal released team-friendly GPU notebooks that start up quickly, with 5,000+ accounts trying them during testing.
ARM created a new Lumex system for smartphones that makes AI run 5 times faster by improving how the computer parts work together.
Google launched AI Quests to teach 11-14 year old students about AI through games about predicting floods and stopping diseases.
Google made a tool called Stax that helps developers test AI models using their own specific requirements instead of standard tests.
TOOLS
Trending AI Tools
🎥 Filmora - Turn scripts or audio into full videos with stock visuals, subtitles & more
⚙️ ChatFast - Build custom GPT chatbots from your own data, no coding needed
🎨 Profile Picture - Generate your ideal AI-crafted profile photo in over 70 styles
🎵 Singify - Generate complete songs from text prompts and lyrics
🎨 ProPhotos - Instantly generate high-quality, photorealistic AI headshots that reflect your style, no studio required
✍️ Niche Traffic Kit - Build traffic-ready niche sites with AI and PLR packs
🚀 Hoop - Automatically capture and organize tasks from Slack, email, and meetings
🎤 Wispr Flow - Dictate across apps 3-4× faster than typing, with voice commands
