# Cal Newport AI takes are WILD...

Source: https://www.youtube.com/watch?v=uWLt81SgM78
Recap page: https://rapidrecap.app/video/uWLt81SgM78
Generated: 2026-03-04T09:05:15.608+00:00

---
## Quick Overview

Cal Newport argues that while AI progress, particularly in coding, is impressive, it does not represent a fundamental, permanent shift in work or intelligence as claimed by some, pointing to the recent slowdown in scaling/inference improvements post-GPT-4 and the continued need for human oversight, as evidenced by the fact that AI labs deliberately focused on coding first, which was a strategy that unlocked other capabilities, rather than the primary goal being general intelligence.

**Key Points:**
- Cal Newport argues that AI progress, especially in coding, has slowed down since late 2024/early 2025, evidenced by the flattening curve on the METR chart after GPT-4o.
- Newport suggests that AI labs deliberately focused on making AI great at writing code first because building AI requires a lot of code, and code-writing ability unlocks broader capabilities, not that coding was the end goal itself.
- He contrasts the exponential hype with the reality of recent progress, noting that the pace of improvement has slowed down significantly post-GPT-4o.
- Newport cites Matt Shumer's article, which claims that AI is now doing the entire job of an engineer (e.g., building an app in four hours), but Newport believes this is an overstatement, calling it 'Grade A nonsense'.
- He points out that the coding AI agents still make mistakes (requiring human validation) and suggests the current focus is on narrow applications, not general intelligence breakthroughs like those claimed by Google DeepMind's AlphaEvolve.
- He highlights the massive, seemingly exponential valuation growth of Anthropic (reaching $18B to $26B in revenue by 2026) compared to OpenAI's growth, noting that investors are nervous about the sustainability of these valuations.
- The video references the Darwin Gödel Machine concept as an example of AI improving itself, but Newport implies this is also overhyped or misleading compared to current capabilities.

![Screenshot at 04:54: The commentator points out Newport's argument that the current rapid progress in AI, especially coding, is not a permanent fundamental shift, contrasting it with the hype.](https://ss.rapidrecap.app/screens/uWLt81SgM78/00-04-54.jpg)

**Context:** The video features a commentator reacting to and analyzing claims made by Cal Newport regarding the current state and future trajectory of Artificial Intelligence, specifically challenging the notion that AI has permanently and fundamentally changed work forever. The discussion centers on Newport's skepticism about the pace of recent progress, particularly in coding, and his critique of the narrative surrounding AI capabilities versus the actual evidence, referencing data from METR and recent announcements from Google DeepMind (AlphaEvolve) and Anthropic.

## Detailed Analysis

The commentator analyzes Cal Newport's skeptical take on the current AI hype, arguing that the progress, while impressive in specific areas like coding, has recently slowed down, contradicting claims that work has changed forever. Newport uses a METR chart showing progress slowing after GPT-4o, suggesting the exponential leaps are over for now. He refutes the idea that AI agents can fully replace engineers (like building an app in four hours), calling such claims 'Grade A nonsense' and noting that these agents still require human oversight and testing. Newport suggests the initial focus on code generation was a deliberate strategy by AI labs because code requires a lot of effort, and mastering it unlocked other capabilities, not that it was the final goal. The commentator also reviews evidence of massive valuation growth for companies like Anthropic (projected to reach $18B-$26B revenue by 2026) despite underlying revenue growth being exponential but still far from the implied generalized intelligence. The speaker concludes that the current situation is not a paradigm shift but rather an overhyped narrative, citing a lack of true fundamental breakthroughs in general intelligence and pointing to the fact that many current AI applications are still narrow.

### Cal Newport's Central Argument

- The AI labs made a deliberate choice to focus on making AI great at writing code first because building AI requires a lot of code, and code-writing capability unlocks everything else, which is a strategy that is now yielding diminishing returns.

### Critique of Hype vs. Reality

- Newport claims the recent rapid progress has slowed down since late 2024/early 2025, and the current claims about AI replacing engineers are exaggerated ('Grade A nonsense') because the work still requires significant human validation and testing.

### Evidence of Slowdown

- The commentator references a METR chart showing the steep progress curve flattening after GPT-4o, indicating that the rapid scaling seen previously has stopped.

### Evidence of AI Utility

- Despite the hype, the commentator notes that AI is being used effectively in niche areas like complex mathematical problems (citing Google DeepMind's AlphaEvolve solving problems previously unsolved) and data center/hardware optimization.

### Financial Metrics

- The commentator highlights Anthropic's explosive revenue growth (projected $18B-$26B by 2026) versus OpenAI's, noting that investors are nervous because this growth isn't yet backed by overall economic impact or widespread enterprise adoption.

### Self-Improving AI Concept

- Newport's critique extends to concepts like the 'Darwin Gödel Machine' and claims that AI is rewriting its own code, which the commentator finds to be exaggerated or not as revolutionary as presented.

![Screenshot at 00:10: The METR chart illustrating the rapid, then slowing, progress curve of LLMs over release date.](https://ss.rapidrecap.app/screens/uWLt81SgM78/00-00-10.jpg)
![Screenshot at 00:38: Cal Newport text snippet stating that AI labs made a deliberate choice to focus on coding first because building AI requires a lot of code.](https://ss.rapidrecap.app/screens/uWLt81SgM78/00-00-38.jpg)
![Screenshot at 01:41: Commentator referencing Matt Shumer's viral article, which he found shocking.](https://ss.rapidrecap.app/screens/uWLt81SgM78/00-01-41.jpg)
![Screenshot at 06:50: The commentator pointing out that the progress curve slows down after the GPT-4o release.](https://ss.rapidrecap.app/screens/uWLt81SgM78/00-06-50.jpg)
![Screenshot at 13:52: A screenshot of Terence Tao's post discussing AI tools solving Erdos problems, which the commentator uses as an example of genuine AI progress.](https://ss.rapidrecap.app/screens/uWLt81SgM78/00-13-52.jpg)