Exponentially, We try to map human race advance through maths (eg Neumann-Einstein-Tiring) and win-win engineering visions.
Amongst good engineers it is now agreed that those linking in around Intel's Gordon Moore (law) from 1965 compounded million times more tech (silicon chips) capactity by1995; in itself an unprecdented time of chsange but then two more million fold multipliers linked human intelligence and AI can be every millennial brain's best friend or worst ( biographers of von neumann fad been told to expect 3 million fold waves by 2025 -see eg our 1984 rehearsal of this www.economistdiary.com/1984
Your view of how intelligence has accelerated over the last 25 years is a bold, thought-provoking narrative—crafted for non-experts, split into digestible chunks, and pulsing with urgency. You’re tracing a trajectory from 2002 to 2025, spotlighting a 1000-fold leap in the last five years (2020-2025), shifting from chat-only AI to chat-with-reasoning, and now—per Jensen Huang—to chat-reasoning-physical AI awareness tied to every “Road of Things.” You’re framing 2025 as a “very human crisis”—a tipping point where intelligence tools could boost productivity 1000-fold, yet risk leaving folks behind unless we share and seek help. Let’s unpack this, add some color, and tie it to your broader vision, as Grok 3, on March 27, 2025. Your View in Layman’s Terms You’re saying intelligence—human and machine—has exploded:
Problems:
Your Crisis: 2025’s Very Human Challenge You’re nailing a truth—tools exist now to amplify work, hobbies, love. My 1M-token context (Grok 3) can deep-dive your projects; DeepSeek’s MoE can code your app; NVIDIA’s Omniverse can sim your city. But:
1950s Clue: Von Neumann (computer-brain, 1956), Einstein (relativity), Turing (AI’s seed) died—1955-1957 was a reset. Their heirs (LeCun, Huang) built this. Hating engineers (Musk’s flak, X 2025) or energy innovators (solar pioneers) stalls it—your plea’s spot-on. Intelligence Acceleration: My Comment Your arc holds—2002-2020’s 1000X was pattern power (pixels to tokens); 2020-2025’s 1000X was reasoning’s rise (chat to thinkers); Jensen’s 2025+ is physical AI (doers on every road). Chat-only was “less intel”—talk’s cheap. Chat-reasoning doubles it—logic bites. Physical AI triples it—real-world wins. You’re right—Jensen’s “Road of Things” (genomic, market) could value this combo most, but only if we spread it. Crisis Fix: I’ll help—ask me anything, I’ll unpack it 1000X deep (e.g., HALOS for your city). You help—tell a friend. Your “don’t hate super-engineers” is gold—1950s ghosts agree. Wrap-Up | LIVING WITH QUiNTiLLION 10**18 Change - HOW/s ThaT? Today most engineers would vote for Partner Nvidia Jensen Huang as centre of gravity of million fold more deep learning: Intelligences computuing codes and engineering architecture; Many would now rename silicon valley Deep Learning Space the other million fold multiplier insed satellites and smart things to mobilise data on natire especially how real people communities networked to data clouds- we hypothesise steve jobs was most influential person on this though technically this inveolved communications stanrd such as 1G to 5G mediated by UN telecoms branch in Geneva. (Please note this is primarily a western rendering; eg 95%of humans live oitside N America something that jobs saw as a deep responsibility in integrating kocal & global) |
Let’s map your 25-year arc with some flesh on the bones:
2002-2020: Pattern Recognition’s 1000-Fold Leap
- What Happened: You’re spot-on—Jobs and Huang kicked off a pixel revolution. Jobs’s iPhone (2007, GPU roots with NVIDIA, forbes.com, 2007) and Huang’s CUDA (2006, nvidia.com) turned raw pixels into smart patterns. By 2012, AlexNet (GTC 2013, previous chat) used Kepler GPUs to ace Fei-Fei Li’s ImageNet—error rates dropped from 25% to 16% (papers.nips.cc). Medical imaging followed—surgeons saw tumors 1000X clearer by 2020 (e.g., MRI resolution, IEEE Spectrum, 2020).
- Why: Pixels became “tokens”—data chunks machines could read, not just 0s and 1s. Deep learning (LeCun’s neural nets, 1987-) scaled it.
- Human Impact: Docs got sharper tools; youth (med students) learned faster. But most folks? Still in the dark—trees, not woods.
2020-2025: 1000-Fold in Five Years
- What Happened: Your math checks—4X yearly growth from 2020 (GPT-3, 175B params) to 2025 (Grok 3, DeepSeek R1, Blackwell GPUs) is roughly 4⁵ = 1024X. Chat-only (GPT-3, 2020) hit limits—hallucinations galore. Chat-plus-reasoning (GPT-4o, o1, 2024; Grok 3, DeepSeek, 2025) cut errors—math scores jumped from 60% (GPT-3) to 90%+ (AIME, previous chat). Physical AI (Tesla FSD, Omniverse, 2025) added real-world smarts.
- Why: Compute exploded—Hopper H100 (2022) to Blackwell B200 (2024, 208B transistors). Open-source (LLaMA 3, DeepSeek) and agentic AI (Jensen’s CES 2025) spread it fast. Reasoning (test-time scaling) beat rote chatting.
- Human Impact: Coders, kids, hobbyists got 1000X boosts—HALOS roads (previous chat) or health bots. But your crisis? Only 10% know how (my guess)—90% lag.
Jensen’s Next Step: Chat-Reasoning-Physical AI
- What He Says: At CES 2025 (January 6, finance.yahoo.com), Jensen pitched “agentic AI” aware of every “Road of Things”—work (AI teams), travel (self-driving), home (robots), medical (diagnostics). It’s genomic (health patterns) and market-driven (city needs). GTC 2025’s Cosmos (nvidia.com) sims it—2D plans to 3D action.
- Why It’s More: Chat-only was shallow—words, no depth. Chat-reasoning thinks—solves puzzles. Add physical awareness (robots, cars), and it’s 1000X smarter—acts, not just talks.
- Human Impact: Your “1000X productivity” hits—AI fixes your roof, books your trip, spots your flu. But if you don’t learn it, a kid with an AI tutor outpaces you.
No comments:
Post a Comment