After a considerable wait, OpenAI has finally unveiled GPT-5. In light of this news, I recently decided to cancel my ChatGPT Plus subscription, which I had maintained since April 2023.
This decision didn’t happen overnight. For months, I realized that ChatGPT wasn’t meeting my expectations. Here’s my experience and what I’ve turned to instead.
My Experience with ChatGPT Plus
As an early adopter of ChatGPT Plus, I immediately subscribed once it was available in my region—and I stayed for over two years.
I remember the AI hype in late 2022 and early 2023. With ChatGPT gaining traction, I felt compelled to learn how to leverage it. The initial excitement was palpable, as ChatGPT could engage in natural conversations, conduct research, draft songs, and help plan my itinerary.
However, the flaws began to surface.
In the early stages, many users encountered inaccuracies that were delivered confidently. At first, I attributed these mishaps to my inability to prompt effectively or the model’s training on outdated data. However, these excuses no longer hold water, nor does ChatGPT.
Is It Me or ChatGPT?
After two years with ChatGPT, I realized: it’s not me, it’s you, GPT.
I’ve mastered the art of prompting AI, but it often fails to impress. Initially, I believed that ChatGPT could replace my writing tasks as a journalist; however, I found its writing to be flat and uninspiring. If you wanted bland prose, ChatGPT was your go-to.
Some of these issues have improved, yet I still spot AI-generated content a mile away—often through awkward sentence structures and overused metaphors. The chatbot continues to falter, now for different reasons.
Though I wasn’t using ChatGPT primarily for writing, I wanted it to help with research and fact-checking. Unfortunately, I was overly optimistic.
When asking questions, I’ve found ChatGPT can confidently provide invented answers. Often, I’ll find it faster to conduct a manual search instead. Countless times, I asked it to confirm an answer I already knew, only to receive completely inaccurate information. These mistakes could have been embarrassing if I hadn’t been vigilant—too many users take chatbot responses at face value.
Realizing that seeking accurate information from ChatGPT was more trouble than it was worth, I turned my attention to its competitors. This shift taught me a vital lesson: relying on multiple tools usually yields the best results.
Alternatives to ChatGPT
My frustrations led me to explore Gemini, which often does a better job of providing source links—something essential for my work. While I still need to verify the information, it generally outperforms ChatGPT in this regard.
In terms of writing, Gemini isn’t much more inspiring than early GPT versions, but it performs better on technical topics. I tested both models on GPU architecture, and Gemini surpassed GPT in accuracy, though it may have been a coincidence.
I’ve now switched to using Gemini for factual research. I know not to take its output for granted, but it generally makes the process easier than with GPT.
I’ve also started cross-checking responses among different chatbots. Admittedly, this can become a tedious cycle, especially when both models provide incorrect data.
A friend convinced me to try Kagi Assistant, which has become my new go-to. This tool aggregates search models and functions well as an alternative to Google Search.
Could GPT-5 Change My Mind?
My primary concern with chatbots like ChatGPT remains their accuracy.
While some prompts generate useful answers, the frequency of inaccuracies makes me wish I had invested more time in traditional research methods. In today’s AI-dominant environment, honing your research skills feels increasingly vital.
Many users find various applications for ChatGPT, from translations to drafting emails and planning projects. While it has its merits, its tendency to fabricate information severely limits its potential.
OpenAI claims that GPT-5 aims to mitigate the inaccuracies I’ve grown frustrated with. The new model is expected to hallucinate less and provide more reliable outputs. It may also confess when it cannot deliver correct information.
I plan to test GPT-5 like I did previous models. If it shows significant improvements, I might reconsider my stance. For now, I’m content with my choice to stop paying for ChatGPT, and I have no regrets.
Feel free to continue exploring content on related topics at Moyens I/O.