Working with data and technology across major industries such as healthcare, energy, finance, and supply chains for over ten years has given Toptal AI developer Joao Diogo de Oliveira a uniquely broad perspective on AI’s real-world uses. Over the past six years, he has honed his focus on AI and machine learning (ML), tackling the field’s most crucial areas: prediction models, computer vision (CV), natural language processing (NLP), and large language models (LLMs) like GPT.

This comprehensive Q&A, encompassing a wide range of topics, encapsulates a recent ask-me-anything-style Slack forum where de Oliveira addressed AI-related queries from fellow Toptal engineers globally. The discussion begins with the most significant current and prospective applications of AI for modern businesses, subsequently delving into more complex AI and machine learning questions geared towards technologists.

Editor’s note: For clarity and conciseness, some questions and answers have been edited.

Grasping the Current and Future Influence of AI

Given your experience, what are the main uses and advantages of AI in the healthcare sectoridentifying antibacterial molecules What does the future of AI in healthcare look like to you?

—M.D., Seattle, United States

AI is already deeply integrated into healthcare. Fortunately (based on my experience), securing funding for healthcare isn’t always an issue, which means there’s substantial potential for future AI innovation. When it comes to newer research endeavors, I am particularly captivated by the use of deep learning for drug discovery (e.g., Though this falls under chemistry, its applications in healthcare will be numerous, and I foresee it significantly advancing the future of humanity. However, I am concerned about the sluggish pace of the many regulations and approval processes in this field, especially compared to AI.

Could you elaborate on the limitations of AI predictive analytics? Which algorithms and technologies do you find most effective for conducting AI predictive analytics and achieving the most accurate estimations?

—M.D., Seattle, United States

That’s an insightful yet challenging question. When it comes to the limitations, I think we need to assess whether something is predictable and if the necessary data is available before attempting to predict it. It’s tempting to assume that AI can predict anything, but sadly, we aren’t there yet. In terms of preferred algorithms, I’m particularly interested in neural networks, but I also believe that decision trees are highly effective when solving particular problems (e.g., regression analysis).

How do you see technologies such as NLP, AI, and CV influencing search engine rankings? To illustrate, what effect does ChatGPT have on SEO?

—M.D., Seattle, United States

My assumption is that in the near future, we’ll witness some astute individuals and companies utilizing NLP, LLMs, and statistics to analyze—and keep tabs on—their competitors. Numerous excellent articles delve into this subject; for instance, this one explores how to monitor your competition with Google Bard. Looking ahead, I anticipate these tools and practices becoming more widely adopted, thereby leveling the playing field.

What are your views on the new AI chip recently unveiled by AMD? Do you think it will revolutionize computing?

—M.Z., Santa Clarita, United States

While it may seem like a mundane response, I don’t believe we have enough data to definitively say whether this chip will truly transform computing. Nevertheless, I was encouraged by the announcement as it introduces competition to other AI chips, and I firmly believe that a monopoly benefits no one.

I’m observing the current AI buzz surrounding how AI will reshape our lives, and it appears to be more than a passing trend, with the potential to accelerate future advancements. In your opinion, what are the fundamental AI concepts that should be taught in high schools?

—K.C., Berlin, Germany

That’s a great question. I firmly believe that we need to start laying the groundwork for teaching AI fundamentals to high school students (or even younger learners). One of the most crucial takeaways for students is understanding that AI is not magical. At least not in its current form. Today’s AI lacks sentience; it boils down to math. If the next generation can grasp the foundations of AI and what lies beneath the surface, they may be less intimidated by it and more inclined to experiment with its capabilities.

Practical Application: Utilizing Artificial Intelligence, Machine Learning, and Large Language Models (LLMs)

As a developer lacking experience in AI/ML theory, what’s the best approach for me to begin incorporating machine learning or artificial intelligence technology into my product development process? Is it naive to rely on pre-built, black box solutions (e.g., Amazon Rekognition or Textract)? Is it worthwhile to invest time and effort in understanding the underlying theory?

—S.L., London, United Kingdom

My advice is to pursue your interests and passions. If you find AI/ML captivating, dive in and don’t solely depend on pre-built solutions or other engineers. Conversely, if you lack the time or don’t envision a future involving AI or ML, then pre-built products are a suitable option, especially given the unparalleled surge in AI tooling over the past six months or so. In short: Choose your endeavors strategically.

How can we seamlessly integrate ML and NLP technologies into Firebase**?**

—B.S., Amman, Jordan

The approach depends on the specific task you aim to address. ML solutions don’t always necessitate high computational overhead. They can manifest as a straightforward regression model with limited iterations (similar to some NLP solutions). Therefore, they seamlessly integrate with Firebase. However, if you’re referring to LLMs, they demand slightly more processing power. While there have been recent developments in this domain (Falcon-7B), you might still want to explore existing APIs or consider building your own.

Is it feasible to enhance an LLM to provide answers in real time (or within a few hours)?

—L.U., Curitiba, Brazil

Yes, it’s definitely achievable. Of course, there will always be some latency, and the larger the model, the longer it takes to generate predictions (or the more GPU resources it demands).

I’m currently working on deploying an LLM model into production. My plan is to create an API for the model using FastAPI and deploy it to Hugging Face or another cloud platform. Are there any alternative options or methods worth exploring?

—D.P., Bengaluru, India

The solution hinges on the project budget. Clients with substantial budgets can readily afford costly GPUs from AWS, while those working with tighter constraints might require developers to assemble a FastAPI and BERT solution designed to run on a CPU in a virtual environment using Vast.ai](http://vast.ai/). The optimal approach depends entirely on the specific [business case and available resources.

Upskilling: Expanding Your AI Development Knowledge

Given that LLMs have started generating code, what essential hard skills should I prioritize learning to remain competitive as a developer and effectively incorporate AI into engineering processes?

—M.M., São Paulo, Brazil

I don’t believe we’ve reached a point where developers are obsolete (though I estimate we could be within 10 to 15 years). Looking at the foreseeable future, I predict that AI might not excel at handling edge cases, customizations, and the numerous special requests clients frequently desire. Therefore, I recommend learning to leverage generative AI to expedite the writing of boilerplate code. Preserve your mental energy for tasks like ensuring the code functions as intended across various scenarios. Instead of dedicating 40 hours to developing a single program, you might find yourself working on 10.

I have four years of experience in computer vision. What courses or skills would you suggest I pursue to transition into working with LLMs?

—M.T.Z., Islamabad, Pakistan

I would advise starting small and initially concentrating on NLP. Once you have a firm grasp of NLP fundamentals, you can delve into LLM nanodegrees offered by online learning platforms to gain an understanding of core concepts like embeddings and transformers. Lastly, I highly recommend experimenting with Hugging Face, which should be relatively straightforward given your background in AI.

Could you recommend any useful resources, tools, frameworks, or sample projects for aspiring AI or ML engineers?**

—A.D.R., Como, Italy

I would suggest two primary resources. Firstly, nanodegrees (online certified programs) serve as an excellent starting point. Stanford Online’s machine learning coursework is highly beneficial if you’re new to the realms of AI and data science. Secondly, to accumulate practical experience and begin experimenting with AI/ML technologies, Kaggle projects and competitions are invaluable resources, offering ample networking opportunities and the chance to learn from fellow enthusiasts.

The editorial team of the Toptal Engineering Blog wishes to express its sincere appreciation to Meghana Bhange for her invaluable review of the technical content presented in this article.