# My worst take of all time

Source: https://www.youtube.com/watch?v=z0dw4z5em2s
Recap page: https://rapidrecap.app/video/z0dw4z5em2s
Generated: 2025-09-27T14:10:31.589+00:00

---
## Quick Overview

The speaker discusses Dave Shapiro's tweet about his "worst take" regarding OpenAI's role in popularizing Chain-of-Thought (CoT) reasoning, admitting he underestimated its importance, and then shifts to discussing the controversy surrounding OpenAI's "Pulse" feature pricing model, contrasting the enterprise focus with consumer needs and the high utilization of GPUs overnight.

**Key Points:**
- Dave Shapiro admitted his tweet claiming OpenAI's Strawberry model lacked Chain-of-Thought (CoT) reasoning was his 'worst take' one year later, conceding he underestimated the development's importance.
- The original CoT paper was published at NeurIPS in January 2022, predating OpenAI's widespread adoption of the technique.
- The speaker analyzes the internal controversy at OpenAI regarding the 'Pulse' feature, where infrastructure teams highlighted that expensive GPUs were idle overnight, prompting finance to question utilization.
- Infrastructure staff reportedly scoffed at the idea of running GPUs overnight because 'users are sleeping,' suggesting a disconnect between enterprise cost focus and consumer usage patterns.
- The speaker notes that despite OpenAI's dominance in the consumer market, their current focus seems heavily weighted towards the enterprise market, which has different needs (like guaranteed service levels).
- The speaker argues that high GPU utilization times, such as 11 AM to 5 PM Eastern Time, are when models should be running, rather than sitting idle overnight.
- The speaker concludes by stating that the core issue is that finance teams often do not understand what a GPU is, leading to poor cost optimization decisions.

![Screenshot at 00:00: Dave Shapiro's tweet stating "Unequivocally my worst take" and referencing being blocked by OpenAI, setting the stage for a discussion on past misjudgments in the AI field.](https://ss.rapidrecap.app/screens/z0dw4z5em2s/00-00-00.png)

**Context:** The video centers on a discussion analyzing two recent tweets by AI personality Dave Shapiro. The first tweet is Shapiro retracting a previous statement about OpenAI's model capabilities, while the second tweet details an internal conversation at OpenAI regarding GPU utilization and pricing for their 'Pulse' feature, highlighting a conflict between infrastructure cost-saving measures and actual user demand patterns.

## Detailed Analysis

The speaker begins by referencing Dave Shapiro's tweet where he admits his previous take on OpenAI's Strawberry model lacking Chain-of-Thought (CoT) reasoning was his 'worst take' a year later, acknowledging he underestimated the significance of CoT, which originated in a January 2022 NeurIPS paper. The speaker then pivots to discussing another tweet from Shapiro detailing an internal conflict at OpenAI concerning the pricing of their new 'Pulse' feature. According to Shapiro's description, the realization that expensive GPUs were idle overnight spurred finance teams to question how to justify the cost. Infrastructure staff reportedly dismissed this by stating users sleep overnight, leading to a conflict where the engineers wanted high utilization (like running overnight) while finance focused on cost amortization and CapEx. The speaker points out that this conflict arises because finance teams often lack technical understanding of GPU usage, leading to poor decisions. He contrasts the consumer market, where usage is high during the day (11 AM to 5 PM ET), with the enterprise market, which requires guaranteed service levels, suggesting OpenAI's focus is shifting too heavily toward enterprise needs, even if it means leaving expensive resources idle overnight.

### Shapiro's Retraction

- Admitting underestimation of Chain-of-Thought (CoT) reasoning importance
- CoT paper published at NeurIPS in Jan 2022
- Strawberry model tweet was his 'worst aged take'

### OpenAI Pulse Feature Controversy

- Finance questioned idle GPUs overnight due to high expense
- Infrastructure staff scoffed at running GPUs overnight because 'users are sleeping'
- Conflict between cost management and actual usage patterns

### Market Focus Shift

- OpenAI currently dominates the consumer market but seems focused on enterprise needs
- Enterprise requires guaranteed SLAs, contrasting with consumer usage spikes during the day (11 AM - 5 PM ET)
- Finance teams lack technical understanding of GPU utilization, leading to poor decisions

![Screenshot at 00:00: Dave Shapiro's tweet admitting his previous take on Chain-of-Thought was wrong.](https://ss.rapidrecap.app/screens/z0dw4z5em2s/00-00-00.png)
![Screenshot at 00:04: The first tweet details Shapiro being blocked by OpenAI and admitting his underestimation of CoT's importance.](https://ss.rapidrecap.app/screens/z0dw4z5em2s/00-00-04.png)
![Screenshot at 00:54: A Perplexity search result embedded in the tweet, asking for the first CoT paper, highlighting the factual basis of Shapiro's claim.](https://ss.rapidrecap.app/screens/z0dw4z5em2s/00-00-54.png)
![Screenshot at 01:36: Speaker points out the timeline where Shapiro was blocked by OpenAI after making critical posts.](https://ss.rapidrecap.app/screens/z0dw4z5em2s/00-01-36.png)
![Screenshot at 02:34: Speaker emphasizes that incremental changes are not enough to satisfy users or capture market share.](https://ss.rapidrecap.app/screens/z0dw4z5em2s/00-02-34.png)
![Screenshot at 03:36: Speaker discusses the need to visit the 'headspace' of the engineers vs. the finance department.](https://ss.rapidrecap.app/screens/z0dw4z5em2s/00-03-36.png)
![Screenshot at 04:37: Speaker mentions that enterprise needs \(like SLAs\) differ from consumer needs.](https://ss.rapidrecap.app/screens/z0dw4z5em2s/00-04-37.png)
![Screenshot at 05:55: The speaker asserts that OpenAI still dominates the consumer market despite enterprise focus.](https://ss.rapidrecap.app/screens/z0dw4z5em2s/00-05-55.png)
![Screenshot at 06:37: The speaker details the infrastructure team's perspective: why aren't we using GPUs overnight?](https://ss.rapidrecap.app/screens/z0dw4z5em2s/00-06-37.png)
![Screenshot at 07:15: The speaker mentions that the majority of users are in America, implying daytime usage patterns are key.](https://ss.rapidrecap.app/screens/z0dw4z5em2s/00-07-15.png)