Blog

                                                                                                                                                                                                   January 29/2024

DeepSeek vs. OpenAI – New Race in AI

Recently, I’ve jumped on the AI ship and started actively exploring how to leverage LLM (Large Language Model) technology in development. To keep things short, here are some highlights of what I’ve learned so far:

     

    • Using Ollama to run LLMs locally on my machine
    • Writing code to interact with OpenAI’s frontier models
    • Understanding the difference between System and User prompts
    • Leveraging summarization, especially for commercial problems or dynamic content using Selenium for JS-loaded pages (scraping webpages based on URL. etc.)

As part of my journey, I’ve extensively used OpenAI’s models, not just for coding but also for everyday tasks like drafting emails and refining my English grammar. However, my curiosity led me to explore a rising competitor DeepSeek, an open-source LLM from China, launched recently, on the day of Trump’s inauguration. DeepSeek seems to have great performance at a fraction of OpenAI’s cost, but its approach sparked questions.

AI generated and badly modified my me.

The first red flag?

DATA Security.
You can’t try DeepSeek without logging in and sharing your email and password, a practice explicitly detailed in their terms and conditions.

https://chat.deepseek.com/sign_in

When you click on Start now on the landing page of DeepSeek you cannot just try it a bit, you are redirected immediately to https://chat.deepseek.com/sign_in.

While many apps collect data, DeepSeek’s origins and the recent TikTok ban in the US amplified my concerns about how user data might be handled.
Interestingly, DeepSeek’s rapid rise as one of the most downloaded apps in the US coincided with TikTok’s ban, deepening my skepticism.

Second red flag?

I noticed significant navigation issues on the DeepSeek website.
For example, while browsing their API documentation at https://api-docs.deepseek.com/, I couldn’t navigate back to the main DeepSeek Platform page.
Instead, I encountered a 503 Service Unavailable error at https://chat.deepseek.com/503/, as shown in the attached screenshot.

https://chat.deepseek.com/503

This lack of seamless navigation and consistent availability raises concerns about the platform’s overall stability and user experience, especially when compared to more polished competitors like OpenAI.
It’s another factor that might deter users from seeking reliable and user-friendly solutions.
It can be said that-

OpenAI is more mature product than DeepSeek.

Comparing the models, I noticed DeepSeek avoids addressing China-related controversies like Taiwan or Tiananmen Square, while OpenAI readily tackles politically sensitive topics.
This raised questions about the trustworthiness of its responses and the possibility of political manipulation.
While DeepSeek’s web interface lacks the polish and user experience of OpenAI’s mature product, the quality of its answers is comparable though my analysis wasn’t exhaustive.

This experience reminded me of the power of perception, much like the Palessi experiment, where people mistook budget shoes for luxury ones. It has proven how people pay more for the brand name.

Palessi experiment, photo from event

Perhaps I’m biased toward OpenAI, but something is unsettling about DeepSeek’s timing and behavior.
As Darwin once said,

“It is not the strongest of the species that survives, nor the most intelligent, but the one most responsive to change.”

As we adapt to these tools, it’s crucial to use them wisely, ensuring they serve us without compromising sensitive information or security.

Although the platform demonstrates strong performance capabilities, it currently supports only two languages and two currencies, which suggests that it’s still a work in progress.
Also, their content / copy is not so clear.

https://api-docs.deepseek.com/quick_start/pricing

My conclusion, for now, is to stick with OpenAI while keeping an eye on how this emerging giant evolves and in which direction it moves.
In parallel, I plan to explore their API documentation and attempt to develop a page scraper using their model, similar to what I’ve done with OpenAI.
This hands-on comparison might allow me to better assess the benefits or simply compare the speed and practicality of DeepSeek’s offering against OpenAI.