Posted on mongodb google news. Visit mongodb google news

<!–

Look at the following. 1) Modify the code to build a chatbot around the functionality 2) I want to store the LLM’s response in MongoDB and save it in flat files. Then, display the response in a readable format on the chat window 3) I must be able to edit the LLM’s response without triggering the multiagent system. Use just the LLM to edit and store the updated content back in the flat files and MongoDB 4) I want to use my LLM (Codestral 22B) stored locally through Ollama instead of OpenAI’s API.

I have been coding throughout this week. Really? Let’s call it “prompting” instead. I have become a prompter instead of a coder. The more intelligent and precise your prompt, the better your code is. You code in C and C++. He codes in Python. She codes in Java. I code in English. Dad was right. I should have taken my English classes seriously.

I am using Cursor AI Pair Programmer. The combination of the Composer functionality with Claude 3.5 Sonnet is a killer. It addresses one of the most significant discomforts in using pair programmers. You scroll through the whole page to determine which parts of your program should hold the bits and pieces of recommendations. The sight of “Your existing code goes here” in the response often gets on my nerves. The Composer’s placement of red and green shades to delete and add code snippets solves the problem elegantly. You just have to click on the “Accept” button.

AI Pair Programmers are at a stage of maturity where the developer still needs to be aware of what happens in the program. Otherwise, the developer can easily get thrown off track. I was sent into a loop of modifications for a couple of hours, during which solving one problem introduced another. Our prompting must be smart enough to break the loop. I don’t imagine a non-developer (or even a citizen developer) breaking free from the divine recommendations easily.

We must know when to stop using the AI Pair Programmer’s suggestions and what they are solving. Many times, the recommendations don’t intend to solve the complete problem. Instead, they introduce debugging statements to uncover more information about the error.

The cursor accesses the complete code base to understand the error and suggests modifications across multiple source code files, including the (dot)env file. We can manually select and add source code files and code blocks to our question. Honestly, after using the “Composer” functionality, I have not looked back at the “Chat” feature. I have yet to explore the “Bug Finder” feature.

The Cursor tool does not carry full intelligence. It is the LLM that runs in the background, which brings in most of the intelligence. We have a choice of LLMs: Claude 3.5 Sonnet, GPT 4o, GPT 4o Mini, o1 Mini, o1 Preview, o1, and Cursor Small. The usage allowance against each LLM is determined based on your subscription category. From the pricing details, I guess that Claude is the best-performing LLM with the Cursor tool.

I am pretty sure there are tens of other functionalities that Cursor and other pair programmers have to offer. I focused only on a few today because I have not explored all of them due to a lack of { }. You end that sentence with anything you want. It will be an excuse.

Is it too much to expect personalised recommendations based on my coding style and commenting practice? The Cursor tool has seen me writing programs for nearly two months now. It was GitHub Copilot previously.

With o3 on the horizon from Sam Altman, which he hints will be the achievement of AGI (Artificial General Intelligence), should developers fear losing their jobs to AI? No. The developer’s role will become more strategic. They will spend more time thinking, designing systems, and planning tasks for AI. Once all three tasks are done, don’t worry; developers will still be required to create the o4, o5, o6, and so on. Happy prompting!

<!–

–>

END OF ARTICLE

Article originally posted on mongodb google news. Visit mongodb google news