ChatGPT, Gemini, and Claude quietly get worse over time. This hidden setting fixes them

For nearly half a decade, the reporter has been engaging with AI-powered language models. During this period, the industry has undergone significant transformations. New competitors such as Gemini and Claude have emerged, while established tools like ChatGPT have undergone substantial evolution. These advancements include enhanced third-party app integration, improved reasoning capabilities, and several other noteworthy improvements.

However, the reporter contends that one aspect stands out above all others: the ability of AI to retain memories. By preserving context from prior conversations, AI becomes more adept at anticipating users' needs within a single session. Additionally, this retention enables the tool to glean more about the user, ensuring that subsequent interactions are personalized to the user's preferences.

Despite these benefits, the reporter has observed a significant drawback to memory accumulation: it can lead to a gradual decline in the AI's performance over time. When starting with a new AI, the reporter has noticed that considerable input is often required to generate responses that meet expectations and fully comprehend the user's profile. Once the AI achieves this "data sweet spot," the user experience improves markedly. However, this positive trajectory often reverses as time passes.