Tuesday, February 25, 2025

TLDR 25Feb25 ... Introduction -- Grok-3 vs competitors, (1) Elon Musk’s xAI adds ‘Big Brain’ reasoning to Grok-3, (2) Microsoft Says It Has Created a New State of Matter to Power Quantum Computers, and (3) Thinking Machines Lab is ex-OpenAI CTO Mira Murati’s new startup

Last update: Tuesday 2/25/25
Welcome to our 25Feb25 TL;DR summaries by ChatGPT of the past week's top 3 stories on our "Useful AI News" page ➡ Introduction: Our  initial results of Grok-3 vs competitors on a complex, non-programming word problem, (1) Elon Musk’s xAI adds ‘Big Brain’ reasoning to Grok-3, (2) Microsoft Says It Has Created a New State of Matter to Power Quantum Computers, and (3) Thinking Machines Lab is ex-OpenAI CTO Mira Murati’s new startup

Podcast  HERE
TL;DR 
 
HERE

ChatGPT's TL;DR summaries of Top 3 stories ...

1. Musk | 2. Microsof | 3. Murati

Introduction   Grok-3 vs its competitors on a complex, non-programming, word problem. Our initial results will be discussed in the first section of the podcast. Details about the extensive "coaching/prompt engineering" required to coax Grok-3/Deep Search towards a valid response will be included in the transcript of the podcast. Indeed, we even modified our prompt to make it easier for Grok-3 to satisfy all of the prompt's specifications.

Of course, this forced us to resubmit the revised prompt to the other six models (2 from DeepSeek) in order to ensure that the responses of all models were comparable to each other. Our initial results suggest that, like many knowledgeable tech students, Grok-3 may be a whiz at Python, but it definitely is not "scary smart" with regards to complex word problems.  

1) "Elon Musk’s xAI adds ‘Big Brain’ reasoning to Grok-3"
-- Jess Weatherbed, 
The Verge, 2/18/25
-- This story also covered by TechCrunchComputerworldBloombergTechDivision (video) 

PROMPT:
Here come text of 2 articles. Provide concise no-nonsense summaries, with key points presented in separate bullets for easier reading. Both summaries should be understandable by savvy computer users.

RESPONSE:

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


2) "Microsoft Says It Has Created a New State of Matter to Power Quantum Computers"
-- Cade Metz, 
NY Times, 2/19/25  
-- This story also covered by WiredEngadgetBloombergGeekWire, and Microsoft (in Nature)

PROMPT:
Here comes the text of an article. Provide two summaries. the first should be concise and written with a touch of irony and wit; the second should be a detailed, no nonsense account with key points presented in separate bullets for easier reading; both summaries should be understandable by savvy computer users. Your summaries should only be based on this article. Do not read any other articles.

RESPONSE:
Summary 1: The Quantum Leap, Microsoft-Style 

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.

BackToTop

Summary 2: No-Nonsense Breakdown of Microsoft’s Quantum Computing Breakthrough

Microsoft’s Bold Claim

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.

How It Works

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.

Why It Matters

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.

The Skepticism

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.

The Big Picture

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.


3) "Thinking Machines Lab is ex-OpenAI CTO Mira Murati’s new startup"
-- Kyle Wiggers, TechCrunch, 2/18/25  
-- This story also covered by ReutersNY TimesForbes

PROMPT:
Here come the text of an article. Provide a concise, no nonsense summary with key points presented in separate bullets for easier reading; the summary should be understandable by savvy computer users.
RESPONSE:
Key Takeaways

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

Core Problems They Aim to Solve

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.

What Thinking Machines Lab Will Build

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.

Big Picture

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.