# The next 36 months will be WILD

Source: https://www.youtube.com/watch?v=zeHTTXAWDUA
Recap page: https://rapidrecap.app/video/zeHTTXAWDUA
Generated: 2026-02-26T14:03:15.695+00:00

---
## Quick Overview

The convergence toward Artificial General Intelligence (AGI) is estimated around the 2027-2028 window, driven by compounding advancements in compute scaling, capital expenditure, and algorithmic efficiency, leading to recursive self-improvement (RSI) and an acceleration loop where models become economically substitutable before achieving consciousness, as detailed by figures like Dario Amodei and Sam Altman.

**Key Points:**
- The '2027 Convergence' estimates AGI/ASI arrival between 2027 and 2028, based on the convergence of compute scaling, capital expenditure, and algorithmic efficiency trends.
- Key AI pioneers like Dario Amodei (Anthropic) and Sam Altman (OpenAI) contribute to this consensus, with Anthropic suggesting AI matching Nobel-level capability in a datacenter.
- The AI development timeline shows NVIDIA (Jensen Huang) projecting competitive ability by 2029, while OpenAI (Sam Altman) projects AGI by 2027-2028, defining 'The Collapse Window'.
- The acceleration loop of Recursive Self-Improvement (RSI) involves five steps: Algorithmic Research, Data Generation & Curation, Writing/Implementation, Training Models, and Evaluating New Models.
- Empirical scaling laws suggest task duration is accelerating exponentially, moving from minutes of coherence for chatbots to multi-week autonomous projects for researchers by 2026.
- The core constraint is the massive 500 TWh AI power demand (around 12% of US 4000 TWh total), necessitating rapid infrastructure upgrades like solar, natural gas, and nuclear SMRs.
- The 'Invisible Displacement' is characterized by a vanishing job ladder, where entry-level roles drop by 13% in AI-exposed roles due to hiring freezes, creating 'ghost jobs' that are never filled.

![Screenshot at 00:18: The slide titled 'THE QUIET CONSENSUS: IT'S NOT JUST HYPE, IT'S THE BUILDERS' visually summarizes the convergence timeline of predictions from Anthropic, the AI 2027 Report, and OpenAI, all centering around 2027 within 'THE COLLAPSE WINDOW'.](https://ss.rapidrecap.app/screens/zeHTTXAWDUA/00-00-18.jpg)

**Context:** This presentation focuses on projecting the timeline for Artificial General Intelligence (AGI) and Artificial Superintelligence (ASI), referencing predictions from leading figures like Dario Amodei and Sam Altman, and introduces concepts like the '2027 Convergence' and the 'Industrial Siege' dynamic driving relentless competition. The speaker utilizes charts to illustrate the exponential growth in AI capabilities (Task Horizon) and the looming energy constraint, ultimately framing the current employment situation as Solow's Paradox 2.0 where productivity rises but job creation stalls.

## Detailed Analysis

The video argues that AGI/ASI is converging around 2027-2028, based on the confluence of improvements in compute scaling, capital expenditure, and algorithmic efficiency, leading to the concept of the '2027 Convergence'. Key figures like Dario Amodei and Sam Altman are cited as proponents of this accelerated timeline. The timeline suggests that by the 2027-2028 window, AI will be capable of recursive self-improvement (RSI), creating an acceleration loop where models quickly surpass human performance in areas like coding (e.g., 14.5 hours autonomous coding projected for Feb 2026). The speaker emphasizes that the development process, RSI, involves five stages: Algorithmic Research, Data Generation & Curation, Writing/Implementation, Training Models (Compute), and Evaluating New Models. A major bottleneck identified is energy demand, projected to hit 500 TWh, requiring massive infrastructure buildout via solar, natural gas, traditional nuclear, and SMRs, creating a '2028 Compute Pause Risk' if energy supply lags. Furthermore, the speaker discusses the economic impact, framing it as 'Solow's Paradox 2.0: The Jobless Expansion,' where productivity increases exponentially (Phase 2: Harvest) while job creation stalls due to AI substitution (Phase 1: Investment). This job displacement is described as the 'Invisible Displacement' or 'Vanishing Ladder,' where entry-level roles decrease by 13% in AI-exposed sectors because the positions are never filled ('Ghost Jobs'). The speaker also introduces an epistemology of risk matrix, categorizing AI risks into four quadrants based on Impact and Predictability, noting that scaling laws (Model Capabilities, Compute Growth) are highly predictable but high-impact, while societal outcomes are low predictability/high impact ('The Zone of Unknown'). The overall message is that the pace of AI progress is dictated by engineering realities that are accelerating, regardless of public philosophical debates about consciousness (Sapir-Whorf Trap).

### The 2027 Convergence

- AGI/ASI estimated between 2027-2028
- Driven by compute scaling, capital expenditure, and algorithmic efficiency
- Key players like Amodei and Altman support this accelerated timeline

### RSI

- The Acceleration Loop: Five steps including Algorithmic Research, Data Generation, Writing, Training, and Evaluating Models
- Models are expected to become economically substitutable before achieving human-like consciousness or soul

### The Energy Bottleneck

- AI power demand projected to hit 500 TWh (12% of US 4000 TWh total)
- Solutions involve solar, natural gas, traditional nuclear, and SMRs, with a 2028 Compute Pause Risk if nuclear/SMR deployment lags

### Solow's Paradox 2.0

- The 'Jobless Expansion': Phase 1 (Investment/Reorganization) shows GDP up but job creation stalled (3.7% GDP growth, 181k jobs created)
- Phase 2 (Harvest) shows productivity explosion and labor decoupling

### The Invisible Displacement

- The Vanishing Ladder: Entry-level roles drop by 13% in AI-exposed roles due to hiring freezes
- Ghost Jobs are positions that are never filled, not mass layoffs

### Epistemology of Risk Matrix

- Categorizes risks by Impact (Low to High) vs. Predictability (Low to High)
- Scaling Laws (Model Capabilities, Compute Growth) are highly predictable but high impact
- Societal Outcomes are low predictability/high impact ('Zone of Unknown')

![Screenshot at 00:18: The initial slide establishes the '2027 Estimate' convergence point for compute scaling, capital expenditure, and algorithmic efficiency.](https://ss.rapidrecap.app/screens/zeHTTXAWDUA/00-00-18.jpg)
![Screenshot at 00:19: A timeline chart showing the 'Collapse Window' for AGI around 2027, with input from Anthropic, the AI 2027 Report, and OpenAI.](https://ss.rapidrecap.app/screens/zeHTTXAWDUA/00-00-19.jpg)
![Screenshot at 02:00: A graph illustrating 'Empirical Scaling Laws' showing task horizon duration \(minutes to years\) accelerating exponentially from 2022 to 2026.](https://ss.rapidrecap.app/screens/zeHTTXAWDUA/00-02-00.jpg)
![Screenshot at 04:06: A diagram detailing the five steps in the 'Recursive Self-Improvement \(RSI\): The Acceleration Loop' cycle.](https://ss.rapidrecap.app/screens/zeHTTXAWDUA/00-04-06.jpg)
![Screenshot at 10:35: A slide titled 'The 'Industrial Siege': The AI Race & Point of No Return' illustrating race dynamics, sunk costs \(trillions\), and the imperative for maximum speed forward.](https://ss.rapidrecap.app/screens/zeHTTXAWDUA/00-10-35.jpg)