DraganSr: 2026-03-20

Friday, March 20, 2026

AI LLM multi-lingual thinking! 联合 (Join)

While using Gemini Flash from Antigravity tool for code generation,
web app strangely failed at one point,
with error that mentioned that 联合 (Join) is missing!?

So I asked Gemini to explain what those Chinese chars came from,
and result was much more fascinating than I ever expected!

Here is the response!
This AI model has very nice manners,
even though it was a bit eager to go back to fixing the code, that is does very well!

AI: NVIDIA Jensen Huang, podcasts

NVIDIA GTC Keynote 2026 - YouTube

NVIDIA Founder and CEO Jensen Huang’s GTC keynote as he unveils the latest breakthroughs in AI and accelerated computing. See how agentic AI, AI factories, and physical AI are powering the next generation of intelligent systems.

Alan Kay's quote, "The best way to predict the future is to invent it"

Jensen Huang: NVIDIA - The $4 Trillion Company & the AI Revolution | Lex Fridman Podcast #494 - YouTube

This conversation with NVIDIA CEO Jensen Huang covers the AI revolution, system engineering, leadership philosophy, and the future of computing.

Key Points

Extreme Co-design: NVIDIA has shifted from just building GPUs to rack-scale engineering, optimizing the entire stack including CPUs, networking, cooling, and software to handle modern LLMs.
AI Scaling Laws: While high-quality human data may be limited, synthetic data and test-time compute (thinking, reasoning, and planning) will continue to drive AI advancement.
Leadership and Strategy: Huang emphasizes reasoning from first principles, testing ideas against the speed of light (physical limits), and maintaining a shared vision within the company and industry.
Supply Chain: Huang expresses confidence in the supply chain, specifically citing a strong, trusting relationship with TSMC built over decades.
Power and Efficiency: Data centers must be engineered to gracefully degrade during power shortages, and utilities need to offer more flexible power delivery.
AGI Timeline: AGI depends on the definition, but Huang focuses on AI accelerating productivity, noting that intelligence will be commoditized while humanity and character remain crucial.

Jensen Huang: Nvidia's Future, Physical AI, Rise of the Agent, Inference Explosion, AI PR Crisis - YouTube @ All-In Podcast - YouTube

This special episode of the All-In Podcast features Jensen Huang, CEO of Nvidia, discussing the future of AI. He outlines a shift from simply building GPUs to creating complete AI factories and agentic systems. Huang emphasizes that Nvidia's roadmap includes disaggregated inference—using a mix of specialized chips to optimize complex AI workflows (1:34).
Key Takeaways

The Agentic Future: AI is moving from simple generation to agentic processing, where agents use tools, access memory, and collaborate to solve complex problems (3:35).

AI Factory Philosophy: A $50 billion datacenter is actually the most cost-effective solution for producing tokens due to extreme efficiency, not just hardware cost (7:33).

Democratizing Computing: OpenClaw is acting as a new operating system for personal AI computers, bringing powerful agentic capabilities to the desktop (15:50).

Physical AI: The next major wave is Physical AI, with a projected $50 trillion market impact across robotics and intelligent factories (12:00).

AI Productivity: Engineers should consume a significant amount of tokens to remain competitive; token usage is the new metric for productivity (24:28).

Jensen Huang Says $500K Engineers Should Use at Least $250K in Tokens - Business Insider

"That $500,000 engineer at the end of the year, I'm going to ask them how much did you spend in tokens? If that person said $5,000, I will go ape something else," he added.
"This is no different than one of our chip designers who says, 'Guess what? I'm just going to use paper and pencil,'" he said, referring to top engineers who underutilize AI tokens.

OpenAI: Guide for building AI agents

a-practical-guide-to-building-agents.pdf @OpenAI

Agents are systems that independently accomplish tasks on your behalf.

Building agents requires rethinking how your systems make decisions and handle complexity. Unlike conventional automation, agents are uniquely suited to workflows where traditional deterministic and rule-based approaches fall short.