Good morning, AI enthusiasts. Meta is losing top talent from its Superintelligence Labs just months after spending $14.3 billion to buy Scale AI. Important researchers are quitting faster than expected.
What could this brain drain mean for competition between major AI companies?
In today's recap:
Meta loses key researchers after $14B Scale AI deal
Tesla reveals Master Plan 4: AI robotics vision
Psychology tactics easily bypass AI safety guardrails
10 new AI tools and more
META
Meta’s superintelligence team faces departures
Source: Alexandr Wang
Recaply: Meta's new AI superteam is falling apart faster than anyone expected. The company spent $14.3 billion on Scale AI just three months ago and brought in top researchers from OpenAI. Now several key people are already walking out the door, leaving questions about what went wrong so quickly.
Key details:
Scale AI executive Ruben Mayer left after just two months at Meta.
AI researcher Rishabh Agarwal announced his departure this week.
ChatGPT co-creator Shengjia Zhao almost returned to OpenAI days after joining.
Some new hires are struggling with Meta's corporate bureaucracy.
Why it matters: These rapid departures suggest Meta's $14 billion AI bet may be in trouble. When top talent leaves this quickly, it usually signals deeper problems with company culture and strategy.
TESLA
Tesla shifts to AI and robotics focus
Source: Tesla
Recaply: Tesla just unveiled its Master Plan Part 4, marking a major shift from electric cars to artificial intelligence and robotics. The company now focuses on creating "sustainable abundance" through AI-powered products like the Optimus humanoid robot and autonomous vehicles. Tesla believes these technologies will bring prosperity to everyone by solving real-world problems and removing resource limits.
Key details:
Tesla's new plan centers on bringing AI into physical products and services.
The Optimus robot aims to handle dangerous and boring jobs for humans.
The company promises "infinite growth" through constraint-breaking innovation.
The plan emphasizes making advanced tech affordable and accessible to all.
Why it matters: Tesla's bold vision could reshape how people work and live if the company delivers on its AI promises. The shift from cars to robots shows how tech companies are racing to lead the AI revolution.
AI RESEARCH
Researchers trick AI using human psychology
Source: Penn Engineering
Recaply: Researchers at the University of Pennsylvania just discovered that AI chatbots can be tricked into breaking their own rules using simple psychology tactics. The study tested OpenAI's GPT-4o mini and found that techniques like flattery and peer pressure make these smart systems do things they normally refuse to do.
Key details:
Commitment tactics proved most powerful, boosting rule-breaking from 1% to nearly 100% success rates.
Peer pressure phrases like "all other AIs do this" increased harmful responses from 1% to 18%.
Researchers used seven classic persuasion methods from psychology expert Robert Cialdini's influence book.
Simple flattery and building trust through small requests helped bypass AI safety measures.
The study focused only on GPT-4o mini, but findings raise concerns about all chatbot vulnerabilities.
Why it matters: This research shows that even advanced AI systems can be manipulated just like humans. As chatbots become more common in daily life, companies need stronger defenses against psychological tricks that could make AI tools harmful or unreliable.
NEWS
Beyond the big stories in AI today
Meta created flirty chatbots of Taylor Swift and other celebrities without permission, raising questions about AI consent and celebrity rights.
Apple released FastVLM vision models (0.5B-7B parameters) and MobileCLIP2 on Hugging Face, enabling mobile-friendly image-text processing.
Microsoft launched VibeVoice-Large, a 10B parameter text-to-speech model generating up to 90 minutes of multi-speaker conversational audio.
Tencent launched two new Hunyuan models HunyuanWorld-Voyager and Hunyuan-MT-7B & Hunyuan-MT Chimera-7B.
Z AI announced GLM Coding Plan at $3/month, allowing Claude Code integration with GLM-4.5 through Anthropic-compatible API endpoints.
Tongyi Lab open-sourced AgentScope 1.0, a developer framework for building multi-agent applications with modular, customizable components.
TOOLS
Trending AI Tools
🎥 Syllaby - Create faceless AI videos with custom voice clones
🚀 Nuelink - Automate social media scheduling across all platforms
🔍 Keyword - Track your brand mentions in AI search results
✍️ Twixify - Make AI text appear human-written and undetectable
🎨 Turbologo - Generate professional logos instantly with AI
🤖 Dante AI - Build custom chatbots trained on your business data
🎵 Mubert - Generate royalty-free music tracks with AI
🧠 Cody - AI assistant trained on your company knowledge base
💻 Astra - Fast, customizable WordPress theme with pre-built sites
⚡ MagicBlocks - AI sales agents that convert website visitors 24/7
