How I Use "AI"

🌈 Abstract

The article discusses the author's perspective on the usefulness of large language models (LLMs) and provides numerous examples of how they have improved the author's productivity in various tasks. The author acknowledges the limitations and potential negative consequences of these models but focuses on demonstrating their current utility.

🙋 Q&A

[01] Usefulness of Large Language Models

1. What are some of the ways the author has used LLMs to improve their productivity?

• The author has used LLMs to help them learn new technologies and frameworks, such as building entire web applications with technologies they've never used before.
• LLMs have helped the author automate many monotonous tasks and one-off scripts, significantly reducing the time spent on these types of activities.
• LLMs have replaced a large portion of the author's web searches for tasks like setting up new packages or projects and debugging error messages.

2. How does the author compare the utility of LLMs to the internet? The author states that the recent advances in LLMs have provided the single largest improvement to their productivity since the creation of the internet.

3. How does the author respond to those who claim LLMs contribute nothing or are just another hype cycle? The author argues that current LLMs are already useful and provides numerous concrete examples to demonstrate this, in contrast to those who claim LLMs contribute nothing or will have no lasting impact.

[02] Limitations and Caveats

1. What limitations of LLMs does the author acknowledge? The author acknowledges the limitations of LLMs, such as their propensity to hallucinate, regurgitate facts, and fail spectacularly due to a lack of robustness. The author also understands the ethical concerns around the training of these models.

2. How does the author position their discussion of LLMs' utility? The author states that they will only be discussing whether current LLMs are helpful to them, and not making any claims about the future capabilities of these models or their potential impact on the world.

3. What is the author's background and how does it inform their perspective on LLMs? The author is a skeptical programmer and research scientist who has spent a significant amount of time interacting with various LLMs over the past year. Despite their background in security research, which has involved demonstrating the failures of AI models, the author has found LLMs to be consistently useful in their work.