Moonshots with Peter Diamandis

Episode #152: The Man Who Invented Prompt Engineering on AI, AGI & The Future of Humanoids
Guests: Richard Socher (“Father of Prompt Engineering”), Salim Ismail
Host: Peter H. Diamandis
Date: February 25, 2025

Episode Overview

In this high-energy episode, Peter Diamandis is joined by AI royalty Richard Socher, a pioneer in prompt engineering and CEO of you.com, alongside futurist Salim Ismail. The conversation explores the rapid pace of AI development, the impact of open vs. closed models, the near future of AGI, the revolutionary potential of AI for science and medicine, and the coming wave of humanoid robotics. With a casual, forward-looking tone, the trio debates hype, substance, and practical challenges, offering guidance for investors, technologists, and anyone amazed to be alive during such a pivotal era.

Key Themes and Insights

1. Rapid Pace of AI Advancement

• The group opens with awe at how quickly Elon Musk’s xAI and Grok 3 were developed, emphasizing the impact of massive capital and elite execution teams.
• Richard Socher confirms that, with $2B, a digital superintelligence could likely be built in 1.5–2 years.
  “With enough resources, you can go hard pretty fast.” – Richard (02:37)
• New models (e.g., Grok 3, DeepSeek, Anthropic’s Claude 3.7) are scoring impressively but hype often outpaces reality.

2. Benchmarking AI Progress

• Peter asks about the relevance of IQ as a metric for AI:
  “Is IQ a relevant measure?” (06:03)
• Richard argues intelligence is multi-dimensional; boiling AI capabilities down to one number or relying on the Turing Test is misleading.
  "The best way to fail the Turing Test is to answer questions so much better than a human could." (06:35)
• Test-time compute and speed are emerging as new forms of “intelligence.”

3. Federated Model Approach and AI Market Landscape

• Richard’s you.com provides access to 40+ models, routing queries to the best engine based on user intent (e.g., programming, medicine, history).
• OpenAI, Anthropics’ Sonnet, and DeepSeek rise and fall in usage, with open models quickly gaining mindshare.
• The team notes recurring fraud attempts—creative uses and abuses are common as model APIs proliferate.

4. Open Source vs. Closed Source Debate

• Richard and Salim agree open-source is catching up, citing history from web servers to telcos.
• "It's undeniable. Open source is gaining... it is very hard to compete with that in the long term.” – Richard (15:15)
• Foundational model companies may become commoditized, like telecoms. True value accrues higher up the stack:
  “You can't build an Uber without Internet everywhere, but Verizon doesn’t get a cut of Uber.” – Richard (17:34)

5. Defining & Achieving AGI (Artificial General Intelligence)

• AGI remains elusive and poorly defined. Richard offers both pragmatic and academic takes:
  • Pragmatic: The automation of 80% of digitized work as a “financial” definition.
  • Academic: Multiple types of intelligence (sample efficiency, social, reasoning) must be measured.
• Physical-world manipulation (robotics) is a facet but not a requirement for superintelligence:
  • “A blind person can be very intelligent. ...So these are not necessary capabilities to have superintelligence.” – Richard (13:46)

[Notable Quote]

“We’re too late to explore the oceans and the world. We’re too early to explore... galaxies. We’re right on time to explore superintelligence.” – Richard (00:24 and 27:56)

6. AI’s Moonshot Impact on Science & Medicine

• Panel is bullish that AI will drive most future scientific breakthroughs.
  • Nobel for AlphaFold is just the beginning; AI will compress “a century of biomedical research into the next five to ten years.” (23:24)
  • Material sciences: AI-powered design prompts (“Design me a superconductor...”) change the game.
• Richard details his team’s work on AI protein design:
  "We synthesized proteins 40% different from natural ones, compared to the Nobel-winning 3%—and they folded and worked." (24:35)

7. AI & Quantum Computing

• Pivotal advances in quantum computing (Microsoft, Google) could unlock entirely new simulation domains for AI, accelerating material and drug design.
  • “If you can simulate a domain, AI can solve almost any problem in that domain. ...It’s just a matter of time.” – Richard (30:18)
• Discussion drifts playfully into multiverse theory and quantum metaphysics.

8. The Race for AI Infrastructure, Energy, and Data Centers

• Is the AI sector overbuilding infrastructure and energy, or are we still short?
  • Salim: Overbuilding, as training gets exponentially cheaper.
  • Richard: Disagrees—efficiency drives more usage (“Jevons paradox”); AI, energy, and data appetite will only grow.
  • Both agree: Real value is not in data center “real estate,” but in unique and trusted applications of intelligence.

9. Trust, Data, and the “Trust Layer” for AI

• you.com’s “trust layer” includes traceable citations, “I don’t know” answers, integration of company data, and user training for better transparency.
  • “When you click on citations, the browser scrolls to the quoted fact... you can quickly build trust.” – Richard (19:16)

10. National Security, Surveillance, and Responsible AI

• OpenAI report on Chinese surveillance tools prompt discussion: Downloading open models doesn’t eliminate the risk of embedded data flows—a constant arms race between privacy, security, and technological progress.

11. Humanoid Robots: Hype, Reality, and Use Cases

• Panel reviews the latest humanoid robots (Clone, Neo’s Gamma, Figure) and debates their utility.
  • Salim: Prefers function over form; why humanoid shapes at all?
  • Peter: “Because it’s cool... I’d feel more comfortable with a humanoid robot in my house” (51:55)
  • Richard: Not zero sum. Some cases will demand custom robots (think seven-armed machines!), others could benefit from humanoids for in-home tasks.

[Notable Moment]

• On the wave of development and fantasy meeting reality:
  “I feel like everyone is working on the AI version of the original Terminator. No one works on a T1000.” – Richard (48:48)

12. Investing in AI Startups & Market Dynamics

• Sky-high valuations (Mira’s “thinking machines” startup at $30 billion) prompt bubble warnings.
• Richard: Best investments have a “virtuous data cycle”—the more you use the product, the better it gets (Tesla, you.com).
• Winning in AI means having industry insight, proprietary data, and user relationships—not just a “differentiated” model.

13. AI Agents & The Agentic Future

• Surge in “agent” development—AI assistants that automate workflow:
  • Marketing: Auto-writing campaigns from new features
  • Journalism: Auto-research & summarize new topics
  • VC Analysis: Auto-extracting and evaluating data room metrics
• Personalized agents (“Jarvis”-style) face technical and social hurdles: privacy concerns, need for deep trust, re-engineering of monetization models online.

14. Crypto, AI Agents, and Economic Infrastructure

• Bitcoin and crypto discussed less as investments and more as potential rails for AI agent commerce.
  • Richard: Crypto is still costly, complex; credit cards are easier for now.
  • Salim: Encouraged by crypto’s growing robustness, but major usability issues remain.
• Michael Saylor’s relentless bitcoin evangelism is both admired and approached with caution.

Notable Quotes & Memorable Moments

• “Don’t bet against Elon.” – Peter (03:03)
• “IQ is a broken metric for AI. ...The best way to fail the Turing test is to be obviously superhuman.” – Richard (06:35)
• “Open source is gaining… very hard to compete with that in the long term.” – Richard (15:15)
• “You can’t build an Uber without Internet everywhere... but Verizon doesn’t get a cut of Uber.” – Richard (17:34)
• “A blind person can be very intelligent... these are not necessary capabilities for superintelligence.” – Richard (13:46)
• “Replicated ten years of antibiotic resistance studies in just 48 hours.” – Peter (23:24)
• “We synthesized proteins 40% different from natural ones… and they had the properties we predicted.” – Richard (24:35)

Important Timestamps

• [00:05] — How quickly could superintelligence be built with enough funding?
• [02:37] — On Elon Musk’s speed and xAI
• [06:03] — AI IQ and measurement limits
• [07:29] — you.com’s federated model approach
• [13:21] — AGI definitions: pragmatic vs. academic
• [15:15] — Open source vs. closed source AIs
• [17:34] — The telco analogy for foundational AI models
• [22:21] — AI's role in accelerating science and medicine
• [27:56] — “Too late to explore the oceans... right on time to explore superintelligence.”
• [29:14] — Quantum computing potential
• [38:49] — Data centers, energy, and Jevons paradox
• [43:49] — Investing in AI startups; importance of virtuous data cycles
• [48:55] — “If you want a musculoskeleton humanoid, have a baby!” vs. robots—debate on humanoid form
• [57:28] — Crypto & AI agents transacting
• [65:17] — Agentic AI: automating white collar workflows

Who Should Listen / Key Takeaways

• Executives and Investors: Practical perspectives on investing and future-proofing in the fast-moving AI space.
• Technologists: Insights on where breakthrough opportunities and challenges lie—AI agents, trust layers, multi-modal models, and robotics.
• General Audience: An inspiring sense of living through an unprecedented, transformative era; the episode is rich with optimism, caution, and a playful tone about the challenges and moonshots ahead.

How to Engage / Learn More

• Explore multi-model search and agentic workflows at you.com
• Follow Peter Diamandis for moonshot innovations and abundance mindset
• Connect with Salim Ismail for exponential thinking and forecasts

Summary prepared for those wanting to grasp the bleeding edge in AI, AGI, and next-generation robotics, all through the lens of those building it.