![](https://cdn.sarvent.net/wordpress/andrewsarver/wp-content/uploads/2023/07/07232758/sarventllc_an_image_of_an_AI_system_generating_code_af6524d7-35e5-4710-9a81-737f6f0f9a22-1.png)
Recently, I have found myself exploring the potential of chat applications using OpenAI’s GPT-4 API. It is crucial to view AI as a tool, as it will produce incorrect responses. Therefore, always verify the information returned before accepting it as accurate.
When I am searching for specific information, I appreciate a direct answer, rather than having to wade through a bunch of stuff I don’t care about.
In the past, I entertained the idea of creating a website that offered straightforward code solutions to common problems. Although this concept never came to fruition, I am now convinced that AI Large Language Models (LLMs) could effectively fulfill this need. A lot of the ones I started will end up in my scripts GitHub Repository.
It is worth noting that leveraging AI comes with associated costs. Last June, for instance, I spent approximately $30 on tokens for the GPT-4 model over about 8 days. This cost will fluctuate depending on the amount of code you request the model to generate versus the amount you correct and then instruct it on what you did, and obviously how much you use it. Despite this, my investment has proven valuable as it has assisted me in generating code for languages I seldom use, thereby accelerating my learning process.
I have discovered that the key to effective AI usage lies in interactive engagement. Rather than supplying it with a lengthy list of requirements and expecting a fully developed app in return, it’s better to engage in a chat session. Break your desired output into separate tasks as well. Due to token limitations, you’ll have to reset the session frequently, which means working in sections is beneficial.
Customizing the system prompt and settings to meet specific needs has significantly improved the readability of my results. For instance, adjusting the ‘temperature’ setting can result in more precise or creative responses. This means a degree of experimentation and largely depends on your familiarity with the language. I also find it more efficient to directly tell the system to “Provide me with code, excluding explanations unless explicitly requested.”
The influence of AI on software development is truly astounding. Tasks that would typically take three to four of my Friday nights, I can now complete in just a few hours. This includes working with unfamiliar languages, which would ordinarily extend the duration of the process. It makes my Kanban process work great.
I wonder when companies will start incorporating this technology for their developers. With the advent of AI, we are presented with an opportunity to accelerate the development process and enhance productivity significantly. The future of software development is indeed promising with AI paving the way for faster, more efficient coding practices.
![](https://cdn.sarvent.net/wordpress/andrewsarver/wp-content/uploads/2023/02/26024641/cropped-asarver_bw_aienhanced-150x150.png)