Addressing the conundrum of imposter syndrome and LLMs

July 28, 2024 12:15 PM

VentureBeat/Ideogram

Join our daily and weekly newsletters for the latest updates and exclusive content on industry-leading AI coverage. Learn More

Imagine you’re driving a car on a beautiful, traffic-free day with cruise control engaged. Your legs are relaxed, and you’re humming along to your favorite tunes. Suddenly, the weather changes, the lights dim and the lanes become less visible. The system prompts you to override cruise control and take manual control of the car. As you start to take action, your mind hesitates, unsure of where to place your leg.

How many times has this happened before? This simple scenario illustrates how our brain functions. If we don’t train our brain, it will take that extra split second to perform the action next time. This concept, known as neuroplasticity, is the brain’s remarkable ability to reorganize itself by forming new neural connections and is fundamental to our cognitive development and adaptability. However, in the era of AI and large language models (LLMs), this natural process faces unprecedented challenges.

The power and peril of LLMs

LLMs, trained on extensive datasets, excel at delivering precise and accurate information across a broad spectrum of topics. The advent of LLMs has undoubtedly been a significant advancement, offering a superior alternative to traditional web browsing and the often tedious process of sifting through multiple sites with incomplete information. This innovation significantly reduces the time required to resolve queries, find answers and move on to subsequent tasks.

Furthermore, LLMs serve as excellent sources of inspiration for new, creative projects. Their ability to provide detailed, well-rounded responses makes them invaluable for a variety of tasks, from writing resumes and planning trips to summarizing books and creating digital content. This capability has notably decreased the time needed to iterate on ideas and produce polished outputs.

However, this convenience is not without its potential risks. The remarkable capabilities of LLMs can lead to over-reliance, in which we depend on them for even the smallest tasks, such as debugging or writing code, without fully processing the information ourselves. This dependency can impede our critical thinking skills, as our brains become accustomed to taking the easier route suggested by the AI. Over time, this can stagnate and eventually diminish our cognitive abilities, much like the earlier analogy of driving with cruise control.

Another potential hazard is the erosion of self-confidence. When precise answers are readily available and tailored exactly to our prompts, the need for independent research diminishes. This can exacerbate “imposter syndrome,” causing us to doubt our abilities and curbing our natural curiosity. Moreover, there is a risk of LLMs summarizing incorrect information based on the context of the prompt and the data they were trained on, which can lead to misinformation and further dependency issues.

How can we efficiently use LLMs without feeling inadequate or running into these risks? In this blog, we will explore the balance between leveraging AI tools and maintaining our cognitive skills. Our aim is to provide insights and strategies to navigate this new landscape without compromising our critical thinking abilities.

Strategies to reduce over-reliance on LLMs

To address this, it’s first necessary to understand the tasks where an LLM is genuinely beneficial and also the ones where its assistance can be too helpful and borderline risky. In this section, we provide practical tips and guidelines on how to leverage these powerful tools to your advantage without compromising healthy learning

Supplement learning and skill development

If you’re learning a new programming language or technology, use an LLM to clarify concepts, provide examples or explain documentation. For instance, I wanted to use YAML configuration because of its readability for my use case. I asked the LLM to provide me with the basic concepts behind the idea I wanted to implement, rather than the direct answer. This helped me understand its structure and the factors to consider while creating the file, enabling me to proceed with my task.

Use it as a starting point to brainstorm solutions for specific use cases when it’s difficult to find exact information online. For example, after struggling to find relevant research articles associated with reducing online model bias for classifiers (most were associated with regression), I prompted the LLM, which provided a comprehensive list of useful pointers and techniques that I could further research in detail.

Using this tool to assist learning can be quite productive and powerful. The natural, conversational-like interaction with the assistant is particularly helpful when learning something new and having follow-up questions about a concept. For instance, I had clarifying questions about cancelable contexts in Golang after reading this blog, which I resolved using ChatGPT.

Strategy: Use the LLM as a tutor to supplement your learning. They can help you understand the technology or approach you are using. Discuss abstract use cases to get better answers. However, practice writing your own code and solving problems yourself to reinforce your understanding and retain new information.

Use LLMs for initial research and inspiration

When starting a new creative project, such as writing a blog post or developing a marketing campaign, use an LLM to gather initial ideas and inspiration. Ask the LLM for a list of potential topics, key points or creative angles. This can help you overcome writer’s block and spark your creativity.

This can also apply to software engineering. If you want to build a new feature but need help with the initial code structure, LLMs are invaluable. For example, I wanted to build an app to disambiguate user questions by asking follow-up questions based on their inputs via Streamlit. I explained the initial implementation structure and asked the LLM for a starting point to build upon.

Strategy: Treat the LLM’s output as a starting point rather than a final product. Use the suggestions to brainstorm and develop your own unique ideas. This approach ensures active engagement in the creative process and prevents feeling like you’re being fed answers. It helps boost productivity by overcoming technical difficulties or writer’s block, allowing you to build upon the initial work.

Enhance, don’t replace, your problem-solving skills

Error logs can be verbose and specific, making them difficult to debug. LLMs can be extremely helpful in this regard. When debugging code, use an LLM to get hints or suggestions on where the issue might lie. For instance, you can ask the LLM to explain a specific error message or outline common debugging steps for a particular problem. Below is an example of how a recent debugging session with the assistant went.

Given the response, I prompted it further to help me identify strategies to improve memory management. This takes us back to our tip of using the LLM to supplement learning. This was the response provided by the bot.

At this point I should have ideally researched the approaches listed by the LLM, myself. For example, I was intrigued by the idea of using the parallel computing library Dask for my use case, however I was tempted to ask the LLM to directly optimize my code using Dask. While it did output the exact function I needed, I didn’t understand how Dask worked under the hood, what APIs it exposed or why the code was faster. The right approach would have been to look through the Dask documentation (or ask the LLM to explain the technology) and attempt to reproduce the function using the library.

Strategy: Instead of relying solely on the LLM to solve the problem, use its suggestions to guide your own investigation. Take the time to understand the underlying issue and experiment with different solutions. This will help you build and maintain your problem-solving skills.

Validate and cross-check information

As LLMs improve at understanding context, they can be effective tools for debating and cross-validating your knowledge. For example, if you’re reading a paper and want to validate your understanding, ask the LLM to provide feedback grounded in the paper. While reading a new paper, I conversed with the LLM to validate my understanding and corrected it where relevant.

Strategy: Whenever you read a new journal paper, blog or article, use the LLM to validate your understanding by prompting it to provide feedback on your comprehension of the material.

Set boundaries for routine tasks

LLMs can be very beneficial for routine, mundane tasks like drafting email responses, simple reports or meeting notes. I’ve also used the LLM to assist with filling out membership application forms that require short bios or motivation statements. Often, I know the content I want to include, and the assistant helps enhance the points I provide. Since it excels at summarization, I’ve also used it for character and word limit application prompts.

LLMs are also extremely helpful for formatting already available content according to a given template, a routine task that can be easily automated with their assistance.

Strategy: Set clear boundaries for when and how you use LLMs. Reserve their use for tasks that are repetitive or time-consuming, and handle more complex or strategic tasks yourself. This balance will help you stay sharp and maintain your critical thinking skills.

Conclusion

LLMs are powerful tools that can significantly enhance productivity and creativity when used effectively. However, it’s essential to strike a balance between leveraging their capabilities and maintaining our cognitive skills. By using LLMs as aids rather than crutches, we can harness their potential without falling into the trap of over-reliance or imposter syndrome. Remember, the key is to stay actively engaged, validate information and continuously challenge your brain to think critically and solve problems independently.

Rachita Naik is a machine learning engineer at Lyft, Inc.

Soham Ranade is a machine learning engineer at Vianai Systems, Inc.

DataDecisionMakers

Welcome to the VentureBeat community!

DataDecisionMakers is where experts, including the technical people doing data work, can share data-related insights and innovation.

If you want to read about cutting-edge ideas and up-to-date information, best practices, and the future of data and data tech, join us at DataDecisionMakers.

You might even consider contributing an article of your own!