1. The Verge – Grok-3 Launch and “Big Brain” Reasoning
• Grok-3 Unveiled: xAI’s latest model, Grok-3, boasts 10x more compute power than its predecessor and claims to outperform OpenAI’s GPT-4o, Google’s Gemini, and DeepSeek’s V3 in early tests.
• New Reasoning Modes: Introduces “Think” mode (shows step-by-step reasoning) and “Big Brain” mode (for complex, compute-heavy tasks).
• Price Hike for X Users: Available to X Premium Plus subscribers ($40/month), with an upcoming SuperGrok plan ($30/month) that offers additional features.
• “Maximally Truth-Seeking AI”: Musk emphasizes that Grok-3 prioritizes unfiltered truth over political correctness, though past versions were criticized for misinformation.
• AI Agent Deep Search: xAI is also launching Deep Search, an AI-powered search engine that aims to be a next-generation alternative to traditional search.
2. R&D World – How xAI Built the “Colossus” Supercomputer for Grok-3
• Massive GPU Cluster: xAI built Colossus, a supercomputer housing 200,000 Nvidia H100 GPUs, inside a repurposed Electrolux factory in Memphis.
• Breakneck Construction: The project scaled from 100,000 to 200,000 GPUs in just 214 days, using Tesla MegaPacks to buffer its 250 MW power demand.
• Technical Challenges: Debugging issues included firmware mismatches, network errors, and cosmic-ray bit flips, making large-scale AI training an “entropy battle.”
• Performance & Competitive Edge: While xAI claims Grok-3 edges out competitors in math, science, and coding, external evaluations suggest only marginal improvements over Google’s Gemini 2.0 and OpenAI’s GPT-4o. AI experts call it a “solid frontier model,” but not a game-changer
-- Cade Metz, NY Times, 2/19/25
- Text NY Times
Microsoft just threw a quantum-sized wrench into classical computing by claiming it has created a new state of matter—because, apparently, solids, liquids, and gases weren’t cutting it anymore. The company unveiled topological qubits, which, if real, could make quantum computing vastly more stable and powerful.
• The Quantum Showdown: Microsoft wants to leapfrog Google’s quantum breakthrough, where an experimental machine solved a problem in five minutes that would take a supercomputer 10 septillion years (yes, that’s a real number).
• The Hardware Hack: While Google and others rely on superconductors, Microsoft is betting on a semiconductor-superconductor hybrid that allegedly makes qubits less error-prone.
• The Skeptics Weigh In: Some physicists aren’t convinced yet, warning that Microsoft still needs to prove these topological qubits actually behave as advertised. After all, quantum systems are notoriously fickle.
• Reality Check: Microsoft has built exactly eight qubits—a long way from the thousands needed to change the computing landscape. For now, their machine is more quantum promise than quantum revolution.
The bottom line? Microsoft’s quantum play is either a historic leap or an elaborate case of “trust us, it’ll work eventually.” Time will tell.
• The company says it has built a “topological qubit,” a new type of quantum bit that could make quantum computing more stable and reliable.
• This is based on a new state of matter, which Microsoft believes could enable breakthroughs in AI, medicine, and cryptography.
• Unlike traditional quantum computers that use superconductors, Microsoft’s approach combines semiconductors and superconductors in a custom chip.
• When cooled to -400°F, the chip exhibits exotic quantum behaviors that could theoretically allow for better error correction—a huge challenge in quantum computing.
• Quantum computers can process massive amounts of data exponentially faster than classical computers, which could revolutionize fields like drug discovery, encryption, and physics simulations.
• Microsoft hopes its topological qubits will reduce the instability that plagues current quantum systems, potentially making quantum computing more practical within years, not decades.
• Quantum experts warn that Microsoft hasn’t definitively proven it has a working topological qubit—its behavior still needs verification.
• Microsoft has built only eight qubits, while meaningful quantum computing likely requires thousands or millions.
• Google’s quantum research has already demonstrated ways to reduce quantum errors with existing qubits, calling into question whether Microsoft’s approach is truly superior.
• Governments and tech companies worldwide are investing billions in quantum computing, recognizing its geopolitical and economic implications.
• If Microsoft’s bet on topological qubits pays off, it could leap ahead of Google, IBM, and Chinese quantum efforts—but it’s still far from proving it can deliver on the hype.
For now, Microsoft is positioning itself as a long-term quantum contender, but as with all things quantum, uncertainty remains baked into the equation.
- Text TechCrunch
• Mira Murati, former OpenAI CTO, has launched a new AI startup called Thinking Machines Lab.
• Mission: Build AI tools that are more customizable, widely understood, and capable than current models.
• Leadership Team:
- Mira Murati (CEO) – Former OpenAI CTO
- John Schulman (Chief Scientist) – OpenAI co-founder
- Barret Zoph (CTO) – Former OpenAI Chief Research Officer
• AI’s inner workings remain a black box: The public and researchers lack insight into how AI models are trained.
• AI customization is limited: Current models are difficult to tailor for specific needs and values.
• AI-human collaboration is underdeveloped: AI should better adapt to different levels of human expertise.
• Multimodal AI systems – Capable of working across text, images, and more.
• Customizable AI models – Designed to adapt to individual user needs rather than offering one-size-fits-all solutions.
• Frontier AI in science and programming – Aimed at driving breakthroughs in scientific discovery and engineerinng.
Thinking Machines Lab wants to push AI beyond current limitations, making it more accessible, transparent, and useful across industries. Expect a focus on AI-human collaboration, advanced multimodal models, and scientific applications
No comments:
Post a Comment
Your comments will be greatly appreciated ... Or just click the "Like" button above the comments section if you enjoyed this blog note.