Robot helping a Data Scientist with his code

Recommended Reading: How I Use “AI” by Nicholas Carlini

Luiz Felipe Mendes
3 min readSep 25, 2024

This reading is for those who are willing to spend a few minutes on a longer post. Since it’s a subject that interests me, I thought it was worth it. I’ll provide a brief summary and some comments below.

Original post: How I Use “AI”

The post contains the opinions and individual experiences of the author Nicholas Carlini, who is a researcher at Google Deepmind. His focus is not on LLMs but on security, specifically “attacks on machine learning systems.”

He begins the text by discussing LLMs in general and mentions that there are two types of people: those who believe it is “just a hype” and others who consider it the greatest change in humanity, predicting that all jobs will be replaced in a few years.

My view is similar. I don’t believe it will be the solution to all our problems, nor that it is the ideal tool for all AI challenges. However, he focuses precisely where I see the greatest contribution: in the productivity of people and professionals.

In the rest of the post, he focuses on showing real use cases in his life where he used tools like Gemini and ChatGPT, explaining why they were beneficial. I found it interesting that he shares the commands, iterations, and generated codes. I’ll cite the topics and cases here, but those who are curious should visit the original post.

Listed use cases:

• Building entire web applications with technology I had never used before.
• Teaching me how to use various frameworks that I had never used before.
• Converting dozens of programs to C or Rust to improve performance by 10 to 100 times.
• Trimming down large codebases to significantly simplify the project.
• Writing the initial experiment code for nearly every research paper I wrote in the last year.
• Automating nearly every monotonous task or one-off script.
• Almost entirely replacing web searches to help me set up new packages or projects.
• Approximately 50% replacing web searches to help me debug error messages.

And some real examples he brought that I found interesting:

  • Creating an application that uses Docker for someone who had no experience with this tool.
  • Researching complex things that are difficult to do with “standard queries” on Google or StackOverflow.

My use cases:

In addition to the cases presented by Nicholas Carlini, I would like to add some things I have done through these LLMs:
• Helping my partner, who has never used LaTeX and doesn’t know how to program, to write her master’s thesis (I wish I had this in my time).
• Reviewing all the emails I exchange.
• Suggesting content for presentations, even creating the skeleton of the slides.
• Translations.
• Answering those basic programming questions, like how to use window functions in PySpark to perform task A or B.
• I also plan to create a web app for my own card game this year, and my idea is to use LLMs and GenAI throughout the entire process.

Conclusions

We are certainly experiencing an AI hype, strongly driven by these image and text generation models. Their quality is truly impressive, even for those who have been in the field and working with it for over 14 years. However, there is a lot of magic and exaggeration from companies and individuals.

It is important for people in the AI and technology fields to understand when these tools make sense to be used and when we have better alternatives. However, I believe that all professionals (and perhaps all people) can use them in their daily lives and work to increase productivity and literally make tedious tasks easier.

PS: I didn’t use an LLM to write this text, but I did ask for it to be reviewed and for suggestions for improvements.

--

--

Luiz Felipe Mendes

Co-founded Hekima, an AI company acqui-hired in 2020 by iFood. Head of Data Science at iFood — Recommendation and Search