I have been contemplating lately. I have a desire to learn about deep learning and TensorFlow, but at the same time, I find myself questioning whether this is the right choice in terms of efficiency and practicality. Ultimately, what I want is to distribute the applications I create and allow people to use them; is it appropriate for me to learn TensorFlow to achieve that goal?

Resources for Deep Learning and Practical Concerns

I feel that developing deep learning models requires tremendous system resources. The costs of high-end GPUs, powerful CPUs, and cloud infrastructure for handling large datasets can be astronomical. However, when I think about it, aren’t companies like Google and OpenAI already providing excellent AI APIs? It might be far more cost-effective and efficient to simply use those APIs to obtain the necessary results and apply those outcomes in my applications. So, is it really worth it to study deep learning and build my own models? This thought process frustrates me and also scares me a bit.

Rapid Development vs. Technical Capability

To find answers to these concerns, I thought from a broader perspective. If my goal is simply to develop applications quickly, then utilizing powerful tools like Google’s AI APIs or OpenAI’s GPT-4 would definitely be more efficient. I can use well-trained models and implement the desired features without the burden of infrastructure investment or learning time, and directly incorporate high-performance models into my projects. This approach can reduce costs and enable rapid development.

So, does studying TensorFlow and deep learning hold no value? There is another perspective on this question. What I gain from studying deep learning and TensorFlow is beyond just the ability to create models. Understanding the basics and principles of deep learning grants me the ability to create my own solutions for unique problems that existing APIs cannot solve. Being able to design and improve custom models to address specific problems, rather than just calling AI APIs, will provide a significant competitive edge.

Furthermore, if I want to develop my own unique applications in the long term, I will need the ability to handle frameworks like TensorFlow directly. Existing APIs will always have limitations. To achieve new features or performance enhancements past those limitations, I must be able to understand and work with the models myself. Ultimately, the path to becoming an AI developer might be about creating and understanding AI, discovering infinite possibilities along the way.

Practical Needs and Technical Competence for the Future

In conclusion, I need to find a balance between practical needs and the technical competencies I will need for the future. For now, it seems prudent to use existing APIs for quick results while also studying the fundamentals of deep learning and TensorFlow to cultivate the ability to create my own models. I believe this is the most realistic and forward-looking approach I can take.

As I prepare for the future, technology will continue to advance, and the cost of systems may decrease, paving the way for a time when high-performance PCs become commonplace. Looking back at the performance and price of the PC my parents bought me in the late 1980s, the world has certainly changed rapidly, and now it is evolving several tens, if not hundreds, of times faster than back then.

On the other hand, there are cases where not all applications require massive deep learning. I think even with a low-spec general server or home PC, it could be possible to build and operate an AI model server for simpler functions.

Conclusion: A Challenge Toward My Own AI

In short, handling deep learning directly is definitely not easy, and at times I will inevitably grapple with the limitations of resources and efficiency. However, if I want to create my own AI and contribute to the world through my own solutions, I want to believe that studying TensorFlow is a challenge worth taking. I hope that by walking my own path, this contemplation and fear can lead to greater possibilities.