CodeMaker AI的动态

??? DevTips Weekly - Issue #2 Programmable CI/CD, Text Classification with Spring AI, 7 Programmers Habits, and Distributed Transaction Management #DevTips #Newsletter

How can you leverage AI text-to-code tools to boost productivity? We highlight how AI simplifies coding tasks within the NOW platform and boosts productivity by up to 50%! Discover more: https://spr.ly/6044cRAhi

The number of addressable use cases by chaining together increasingly advanced AI components is only going to grow. For example, combining AssemblyAI with an inexpensive LLM like OpenAI GPT 3.5 via Langchain can lead to very customizable summarization or contextualization of a conversation. Yes, this is 'connecting pipes' on one level. On another level, it's a new category of systems design, where workable definitions of reliability, scalability, and reproducibility are significantly different from a 'traditional' app. As an example of how generative AI demands new ways of thinking about the integration between components, take this paper: https://lnkd.in/e7NN79zn Here, they're designing a new interface (Agent Computer Interface, inspired by HCIs) to enable the agent to be more effective at SWE tasks. These kinds of 'purpose built' interfaces (rather than tying LLMs into UI/UX that is meant for humans) will continue to emerge, and play a large role in the shape of systems leveraging generative AI moving forward. Always interested in developments in this area - if you want to chat, reach out.

AI code bots like GitHub Copilot are useful, yet imperfect tools, and need to be guided and constrained by prompts in order to make them effective pair programmers. But this is exactly the point of a unit test in a TDD environment, except that the computer is the programmer writing the implementation, and you are the programmer writing the test. #AI assisted #TDD really does make allot sense, but it’s wise to use your brain too. https://lnkd.in/gVhvqQvw

"AI can write code way faster than a human programmer, it does such a poor job that making the code useful actually makes it far less efficient than getting a qualified human to just do the job in the first place." A research paper from Princeton University and University Chicago stated the "best-performing model, Claude 2, is able to solve a mere?1.96% of the issues." Jason Hall of Pirates Software fame “We have talked to people who’re using AI-generated code, and they are like, hey, it would take me about an hour to produce this code and like 15 minutes to debug. And then they are like, oh, the AI could produce it in like 1 minute, and then it would take me like 3 hours to debug it. And they are like, yeah, but it produced it really fast.” https://lnkd.in/eAP8wUDB

Evaluating LLM to generate unit testing. Is it already worth it? This article presents an experiment for analyzing different LLMs for generating unit tests. The results are not very promising. As a community, we have the opportunity to build on these experiments, collectively generating more knowledge on where and how we can use these tools to enhance developers' experience and efficiency. https://lnkd.in/dqd_5Vm2 #AI #developerexperience #softwareengineering

Explore how Luxoft utilizes Generative AI and Large Language Models (LLMs) to transform application development. Our latest blog covers everything from addressing common LLM challenges to applying effective solution patterns. Ready to master your AI expertise? Read the full article below. ? https://lnkd.in/dc3qqcan

?? Master Generative AI Plugins with Spring AI ?? (100% my writing, without ChatGPT) Want to build generative AI plugins but don't know where to start? ?? Check out my latest guide on using Spring AI to create and integrate powerful AI functionalities into your Java applications. This step-by-step walkthrough is perfect for developers eager to harness AI in their projects! ?? Read Part I here: https://lnkd.in/drMYbeg2 #GenerativeAI #SpringAI #Java #MachineLearning #AIPlugins

5-Part Post: Building a Document Storage System with AI Tools Part 4: Overcoming Challenges and Building the Foundation The biggest challenge was dealing with software compatibility issues, especially in Python. Libraries often had conflicting dependencies, and debugging these issues manually would have been time-consuming. Fortunately, AI tools like DeepSeek and ChatGPT provided solutions, suggesting compatible libraries and helping refactor the code. However, while AI was invaluable in handling much of the grunt work, it wasn't a complete solution. I still relied heavily on traditional methods like Stack Overflow searches and trial and error to resolve many issues. While AI excels at generating complex methods and assisting with specific coding tasks, attempts to have it write the entire application at once proved unrealistic. Furthermore, AI models can sometimes hallucinate, leading to unproductive cycles and, when using cloud-based AI services, potentially increased costs. Refactoring with AI also requires careful scrutiny, as it can make assumptions and not always adhere precisely to instructions. In essence, AI acts as a sophisticated code-generating tool, much like a dynamic Stack Overflow that directly integrates code, significantly speeding up development. However, thorough code review remains paramount, as neglecting this step has consistently led to problems. Within a couple of days, I had a solid foundation for the project, which I hosted on GitHub (document-storage-system). While the project is ongoing, it already demonstrates the power of AI-assisted development.

It's fascinating to see what Gen AI can be used for and not in the development space to get someone up to speed in a new context.