Julia Community 🟣

Cover image for GenAI Mini-Task: Buy More Time by Mastering Mini-Tasks
Jan Siml
Jan Siml

Posted on • Originally published at siml.earth

GenAI Mini-Task: Buy More Time by Mastering Mini-Tasks

#generativeai #genai #prompting

TL;DR

This blog highlights how Generative AI can be a time-saving asset for Julia programmers, offering an economical way to 'buy' time by efficiently handling everyday mini-tasks and minimizing distractions.

Welcome to the Future of Programming!

Hello, Julia enthusiasts! We're kicking off an exciting new blog series exploring how Generative AI, especially through PromptingTools.jl, can revolutionize your coding experience - "GenAI Mini-Tasks". Imagine turning the mundane, time-consuming tasks into a breeze. That's what we're here to uncover!

The Magic of Mini-Tasks

Mini-tasks: those pesky, everyday chores like reformatting text, tidying up messy files, or coding a utility function for data plotting. They often trip us up, yet they are unavoidable to reach our main goals. What if we could delegate these to a smart, efficient helper?

Why Delegate to AI?

Economically Sensible: Save those precious minutes for just a few cents. It's not just about saving time; it's about investing it wisely in things that actually matter.

Cognitive Load: Every time you pause your main task to tackle a mini-task (eg, "how do I ..."), you're draining mental energy. Generative AI steps in to take that load off your shoulders, allowing you to focus on what truly matters.

Learn As You Go: Engaging with AI tools is a fantastic way to understand their capabilities and limitations, fine-tuning your approach to get the best results. This will become the new way of working/interacting, so why not get ahead of the curve?

The Hidden Cost of Switching

Did you know...

Research shows that context switching is more costly than we realize. Each interruption can set you back >20 minutes in regaining focus (I recommend reading Attention Span). It seems that once we're interrupted, we are more likely to switch to another task, and another, and another... only to come back to the original task (eg, VSCode->Chrome->Email->Slack->Teams->Email... sounds familiar?).

This is why PromptingTools.jl tries to keep you in the REPL!

Real-World Example: String Manipulation

A task that came up recently:

  • You're given a list of currency pairs in the format "BTC/USDT"
  • You need to change the double quotes to single quotes (eg, "BTC/USDT" -> 'BTC/USDT')
  • You need to flip the order of pairs (eg, 'BTC/USDT' -> 'USDT/BTC')

This is easy in Julia, right? Yes, but it costs more time and mental energy than it seems.

Let's practice delegating mini-tasks to GenAI!

using PromptingTools

# Example data to keep example smaller
pairs = """["BTC/USDT", "ETH/USDT", "BCH/USDT", "XRP/USDT", "EOS/USDT"]"""
Enter fullscreen mode Exit fullscreen mode

Let's use a template prompt DetailOrientedTask with placeholders {{task}} and {{data}} and split the tasks into two parts to make it easier for GenAI:

# Set temperature=0 to avoid creative responses
msg = aigenerate(:DetailOrientedTask; task="Change the double quotes to single quotes. Flip the order of currencies", data=pairs, api_kwargs=(; temperature=0))`
Enter fullscreen mode Exit fullscreen mode

It took ~30 seconds to write the command and we get solid results back in ~1 second.
The cost is two-ten-thousands of a cent (0.0002 USD)!

[ Info: Tokens: 121 @ Cost: $0.0002 in 1.1 seconds
AIMessage("['USDT/BTC', 'USDT/ETH', 'USDT/BCH', 'USDT/XRP', 'USDT/EOS']")
Enter fullscreen mode Exit fullscreen mode

I can practically hear: "But that's a toy example! I can do it manually!"

Well, the full task was 300 pairs and it wasn't that much slower/more expensive:

[ Info: Tokens: 2174 @ Cost: $0.0034 in 6.1 seconds
Enter fullscreen mode Exit fullscreen mode

So we have our results in a matter of seconds for a fraction of a cent.

I hear: "But this task is super easy to code up in Julia!"

Assuming that something is “easy” is the most common mistake. I didn't think twice and started writing a list comprehension, then I realised I need to split twice, then I need to remove the square brackets, and then add them back elsewhere, then I need to join, then I need to add quotes, then I need to flip the order, then I need to join again, then I need to join the list of pairs... and I keep writing and re-write code and I'm not done.

8 minutes later I was done with a Frankenstein-code that I'm not proud of:

pairs_flip = []
for pair in split(pairs, ",")
    temp = replace(pair, "\"" => "")
    currencies = split(temp, "/")
    currency1 = strip(currencies[1]) |> x -> replace(x, "[" => "", "]" => "")
    currency2 = strip(currencies[2]) |> x -> replace(x, "[" => "", "]" => "")
    push!(pairs_flip, string("'", currency1, "/", currency2, "'"))
end
pairs_flip = join(pairs_flip, ", ") |> x -> string("[", x, "]")
Enter fullscreen mode Exit fullscreen mode

It would have been much easier to do some of the cleanup in the editor to have easier types and then use Regex substitutions to do the rest, but I didn't think about it because it was "easy". Also, I'm not that familiar with Regex substitutions, so I would have had to Google it anyway -> context switch! Lastly, this has no comment or documentation, so I would have to add that to explain what I'm doing.

Summary: GenAI version is easier to read, faster to produce (30s vs. 8 minutes), and it costs a fraction of a cent. It's an excellent productivity boost if the value of your time is higher than $0.03/hour (0.0034/7.5*60 ≈ $0.03)

Embracing AI in Your Workflow

So, there you have it! Generative AI isn't just a tool; it's a productivity partner. Stay tuned for more insights and use cases in our upcoming posts. Dive into the world of AI-powered programming and experience the difference in your Julia projects!

The credit for the cover image goes to DALL-E 3.

Top comments (0)