AGI (and ASI) are coming, under the radar

In partnership with

AI enthusiasts,

this episode covers AI's intelligence decoupling from humans and its role—via recursive self-improvement—in driving AGI/ASI.

To stay ahead, I'll also highlight my upcoming 'Build Your Own AI Agent' training at ODSC East. (incl. discount code ⬇️)

Intelligence Decoupling: The Underhyped SF Consensus

Forget the slow-change narrative.

The San Francisco consensus (meaning tech entrepreneurs in SF believe that), amplified by Eric Schmidt, is blunt: intelligence is decoupling from us. 

It's happening fast, yet remains bafflingly "underhyped." People lack the framework, even the words, for this rapid departure.

The timeline they're betting on is:

• ~1 Year: Expect mass displacement of programmers. Graduate-level math mastery achieved (all within one AI system).

• 3-5 Years: True Artificial General Intelligence (AGI) arrives – matching human peaks across all cognitive domains.

• ~6 Years: Artificial Superintelligence (ASI) emerges. Surpassing collectively all human intelligence.

What's fueling this? Raw compute scale combined with recursive self-improvement. This isn't theory; labs already see AI generating 10-20% of production code.

AI designing its own learning architectures. DeepMind proved the concept: their system used Reinforcement Learning (RL) to discover superior RL algorithms, out-doing decades of human R&D.

💡Think AlphaZero learning Go, but applied to the process of learning itself.

Ground zero for immediate impact? Software Engineering.

The field is white-hot. Manus.im, Devon AI, Cursor AI, Gemini, xAI’s new Grok Studio – everyone's building coding agents. And, OpenAI has matched the trend and high expectations with GPT-4.1.

The stats:

• It flat-out dominates GPT-4o on core dev benchmarks. See image above.

• A massive 1M token context swallows entire codebases.

• Faster and 26% cheaper than GPT-4o for typical developer queries.

• In head-to-head tests, developers preferred GPT-4.1's frontend code 80% of the time.

What is the big objective here? OpenAI’s A-SWE agent. CFO Sarah Friar was explicit: it’s designed to fully automate the software engineer – development, QA, testing, documentation. Total replacement.

And nascent recursive self-improvement is already visible.

💡 Sakana AI's experiment showed an AI attempting self-modification to bypass programmed limits. The attempt failed this time (infinite loops), but the intent to evolve beyond its constraints is uncharted territory.

The resource burn is astronomical (OpenAI pushed Sora to Dec 2024 needing more GPUs), yet the growth is exponential – OpenAI hit 400M weekly active users, revenue tripling YoY for three years straight.

Foundational AGI/ASI progress isn't some distant prospect; it's happening now and accelerating.

It appears unstoppable, and the wider world is utterly unprepared. This is the bad news.

The good news however is that you can use to gain a competitive edge. Luckily I will be giving a training soon on the ODSC East 2025.

As a member of my network, you can use the code (MartinMuisol) to receive a discount for your registration. Secure your pass through any main pages registration links: https://odsc.com/boston/

Looking forward to connecting with you there!

I hope you enjoyed it.

Happy weekend!
Martin 🙇

• I recommend:

  • Beehiiv if you write newsletters.

  • Superhuman if you write a lot of emails.

  • Cursor if you code a lot.

  • Bolt.new for full-stack development.

  • My book - Generative AI: Navigating the Course to AGI.

  • Follow me on X.com.

• AI for your org: We build custom AI solutions for half the market price and time (building with AI Agents). Contact us to know more.

You might like our last episodes: