After CompSci

Want to become 30% more efficient with your work? Either for your day to day or side projects, the AI tools of 2024 can help. They do require some upfront knowledge so think of their use as the same as steroids, it builds on top of your maximum output that you currently have. Here’s some of the tools I’ve been using.  

ChatGpt

So far, I’ve found three main ways to interact with ChatGpt. 

I have been using it as an enhanced search mechanism. This is for docs or things that I have tangential knowledge of. This has been useful for things that I have a bit of knowledge on so that I can ask the right questions to the model. It’s useful for describing concepts too. In particular, for front end programming, it’s been great since I’m not a high level frontend dev. This has been helping me knock out simple projects. Also, you can get some code scaffolding for the questions you ask here.

You can also generate ideas and brainstorm with it. It can help with writing prompts or rephrasing content. You can have the model mimic many different kinds of people to brainstorm with or to test out content on.

If you need an image generated, DALLE can do that for you via ChatGPT. If you want a specific design, you can do that, too. I do that by uploading an image and then asking the model to describe that image’s style. Then, I take the output description and an input image. I tell the model to generate an image in the style of your output description. So far, this has worked pretty well. 

OLLAMA

I use Ollama for running models locally on my machine. These allow you to run free models such as llama from Meta or Minstral. You can download it here: https://github.com/ollama/ollama. This is great if you don’t want to access the internet after you have the model installed or as a ChatGPT alternative. If you want to create images you will need to run something like llava.

At this point, it’s amazing that if you have some compute and a few gigs of storage, you can run these models locally. Take a look at all of the models that they offer.

GitHub Copilot

There are two versions of this. There’s the advanced type ahead, which is fantastic. It can even create whole blocks of text or entire blocks of code if you describe it. Also, this helps with comments, which can be hard to write efficiently. Let’s be real: Writing isn’t necessarily a trait most engineers are known for. Overall, I have liked this since VSCode is by day to day editor.

Vercel V0

Describe the front end you want, and it will create it for you—full react with tailwind. You can then commit the code, and Vercel will host it for you. This was an excellent move for them to cement themselves as a hosting provider. 

Replit

Another prototyping tool. They used to host environments so that you could just dev. Now, they made it so that you can use an AI assistant when prototyping. The main differentiator with them is that they host everything, from your development environment to the code that you are shipping.

Cursor

It feels like the next level of Copilot. There’s multiline completion. You can hit cmd + K and get it to generate a piece of code for you. Also, it allows you to ask about portions of code that were written to gain understanding, as well as all built into the IDE itself.

The landscape is shifting in this tool space monthly, if not weekly, so keep your eye out. Remember that the value that we add to engineering is not just by the code we produce but by what problems we can solve with the code that we produce. So, if we can get more efficient at creating solutions for the problems, it’s a plus.

Happy Building!