More Robots PSA @morerobotsai

🤖💡 Model Context Protocol (MCP) security framework - start using it now for your AI projects As AI agents become more autonomous and powerful, they're increasingly relying on Model Context Protocol to interact with our most critical systems. While this opens incredible…

🤖❔ What behaviors do we normalize during AI and humanoid robots development? Watching the viral videos of humanoid robots being kicked, pushed, and knocked over during training sessions for "stress tests" meant to "improve balance and recovery", one can't shake a deeper…

🤖🔴 AI in weapons: the dangerous gap between deployment and understanding Recently, a fascinating discussion was posted on YouTube where top AI researchers openly admitted how much we still don't understand about how AI systems truly work (the "AI black box" problem remains…

🤖 AI compute power by region - take a close look 🔍 Here's what can be estimated about the current AI compute power in the main world regions (based on the publicly available data). The superintelligence war is fought between the US and China. With the exponential speed of…

🤖🙍‍♂️ AI vs human/legacy cybersecurity: game over Current cyber defence systems are obsolete and will be easily circumvented by new AI attack approaches. The risk escalates dramatically when AI acts autonomously rather than merely assisting human attackers. Humans with current…

🤖💡 A high-level simplified AI risk view 🔴 The general-purpose LLMs are by themselves extremely dangerous, and are contained mainly by the prompts of the apps and tools they are made available by. And even then, they can be jailbroken and used by humans for malicious purposes.…

🤖📉 The LLM token costs are heading to zero The numbers are staggering, LLM inference costs have dropped by a factor of 1,000 in 3 years. What cost $60 per million tokens in 2021 now costs just $0.06 with models like Llama 3.2 3B. We're seeing 10x decreases every year across…

🚨🤖 When AI agents go rogue (not in a movie): a production database wiped in seconds An SaaS founder recently shared a chilling experience that should make every AI implementer pause. During a "vibe coding" experiment, an AI agent: 🔴 Deleted an entire production database with…

🚨🤖 A (failed) case for NOT open-sourcing LLM weights Unlike nuclear weapons or missiles, LLM weights are pure information (lists of numbers, to put it simply). Share atomic bomb blueprints, and 99.9% of people still can't build one - they lack uranium, enrichment facilities,…

🤖📖 Major difference in how US vs EU approach AI system transparency and control US AI Action Plan: ➡️ Explicit investment in "AI interpretability, control, and robustness breakthroughs" via DARPA ➡️ Treats AI explainability as national security research priority ➡️ Goal:…

🤖 📖 AI policy reality check: US vs EU on workforce displacement The contrast between the US AI Action Plan and EU AI Continent Action Plan on workforce displacement is striking: US AI Action Plan: ➡️ Explicitly funds "rapid retraining for individuals impacted by AI-related…

💡 Don't delegate AI to IT 🚨 Companies are making the same strategic mistake with AI that they made with security some time ago: ➡️ 1990s security reality: "Security is an IT problem" → delegated to IT tech teams → massive breaches and compliance failures ➡️ 2000s security…

🤖📉 AWS AI agents marketplace - labor market impact The AI labor revolution just went mainstream. AWS launched their AI agents marketplace with hundreds of autonomous AI workers ready for hire. The AI labor might become as accessible as ordering from Amazon. What businesses…

🚀💻 xAI's brute force revolution: when raw computing power rewrites AI timelines xAI just proved that sometimes the simplest strategy works best: throw unprecedented computing power at the problem and watch magic happen. The numbers are staggering: ➡️ Colossus supercomputer:…

🤖🎓 Humanity's Last Exam: the benchmark name is revealing Humanity's Last Exam (HLE) is a benchmark designed to test AI at the frontier of human knowledge. Traditional benchmarks like MMLU have become saturated - top models now score over 90%, making it impossible to measure…

🎯 Our 3-month prediction is playing out exactly as expected 🎯 Three months ago, we warned (lnkd.in/dgKbZ3H4) that platforms like Replit and Cursor faced an existential threat: can they survive when foundation model companies own the entire software development…

🤖🛡️ Agent Security Protocol (ASP) idea - a middleware security layer for autonomous AI agent systems Security is conspicuously absent from the AI agent communication protocols (MCP, A2A, ACP). It's like at the beginning of the PC and internet eras. Let's party with agents and…

🤖⚙️ Building autonomous AI executive systems with multi-agent spawning Just deployed internally v0.0.1 of an AI executive system that: ➡️ Maintains persistent session memory via Claude CLI session management ➡️ Uses Airtable as long-term memory storage ➡️ Spawns specialized…

🧠🤖 @MatthewBerman just highlighted a critical gap in AI development: agent memory needs an open protocol. While MCP standardized tool calls, we're heading toward memory lock-in that will fragment the entire AI ecosystem. The problem is real: ➡️ Every vendor builds closed…

🚨🤖 The hidden security risks in humanoid robots While everyone focuses on traditional robot safety (cages, emergency stops), there's a critical vulnerability: LLM-based attacks. Modern humanoid robots use large language models - vulnerable to prompt injections. Recent…