🕵🏻‍♂️X's Grok Expansion

X set to deepen Grok AI integration, app researcher reveals

Key Points 

• App researcher Nima Owji has found code on the website indicating that X plans to roll out at least three new Grok-powered features across its platform.

• These upcoming Grok-related features are likely to be paywalled, similar to the current setup.

🤖News - Elon Musk's X is looking to further integrate xAI's Grok chatbot into its platform.

App researcher Nima Owji has found code on the X website that reveals upcoming plans for the chatbot, indicating Musk's growing reliance on AI for the platform. While these features are not yet publicly available, Owji's findings provide a preview of what's to come.

📱Expected features - As per Owji's findings, X plans to roll out at least three new features across the platform that are powered by Grok.

1. X will reportedly enable subscribers of X Premium and Premium+ to highlight text on the platform and query Grok about the selection through a pop-up prompt without leaving the page. Owji discovered code within X that triggers an "Ask Grok" pop-up message when text is highlighted.

2. Based on Owji's research, another upcoming feature on X will allow users to access Grok without navigating away from the page via a "More about this account" prompt. This feature will allow paying users to request background information about a specific X account directly from an X user's profile card using a Grok button.

3. Furthermore, X plans to introduce a "Grok drawer" allowing subscribers to access the AI chatbot window from anywhere on X. This feature ensures Grok remains accessible and functional as users navigate throughout the site.

It's important to note that these upcoming Grok-related features will likely be paywalled, just like the current Grok AI setup.

Coding with ChatGPT: Success Rates, Errors, and Fixes

Key Points 

• ChatGPT's ability to generate functional code varies widely (0.66% to 89%) depending on factors such as task difficulty and programming language.

• ChatGPT excelled on pre-2021 LeetCode challenges ChatGPT excelled on pre-2021 LeetCode challenges, likely due to limited training exposure.

• ChatGPT can handle syntax errors well but struggles to correct deeper logic mistakes without clear understanding.

• Developers should provide additional context to enhance ChatGPT's understanding and prevent vulnerabilities in complex programming tasks.

👨🏻‍💻Context of the news - For many years, programmers have been writing code for AI models. Now, it's come full circle—AI is being used to write code itself. This prompts the question: how good is AI at coding compared to human programmers?

📔News - A study published in the IEEE Transactions on Software Engineering looked at code generated by OpenAI’s ChatGPT. It found that ChatGPT can produce functional code with success rates ranging from as low as 0.66% to as high as 89%, depending on how hard the task is, what programming language is used, and other factors.

👨🏻‍💼An expert's opinion - Yutian Tang, a lecturer at the University of Glasgow involved in the study, explained that using AI to write code could help make work more efficient and automate tasks in software development. But it's important to understand what these AI systems are good at and where they might struggle.

To dig deeper into these issues, Tang's team tested how well GPT-3.5 could handle 728 coding challenges from the LeetCode platform in five programming languages: C, C++, Java, JavaScript, and Python.

🔎Findings -

• ChatGPT effectively solved coding challenges that were on LeetCode before 2021, achieving high success rates (89% for easy, 71% for medium, 40% for hard problems). However, when faced with algorithm problems introduced after 2021, ChatGPT's ability to produce correct code is affected, resulting in significantly lower success rates (52% for easy, 0.66% for hard problems), likely due to lack of exposure during training.

• The researchers also looked into whether ChatGPT could fix its own coding mistakes after getting feedback from LeetCode. They chose 50 cases where ChatGPT initially made errors, usually because it didn't understand the problem well. While ChatGPT could correct simple coding errors, it had trouble fixing its own mistakes. Tang explains, "ChatGPT might write wrong code if it doesn't understand the problem's logic, so basic feedback alone isn't enough."

• The researchers found that code generated by ChatGPT often had vulnerabilities, like missing null tests, although many of these issues were easy to fix. They observed that code generated in C was the most complex, followed by C++ and Python, which was similar in complexity to human-written code. Tang advises developers using ChatGPT to give it more information to understand problems better and avoid these vulnerabilities. For example, when dealing with tougher programming challenges, developers can share extra knowledge and highlight potential issues for ChatGPT to watch out for.