AI Memory Evolution: ChatGPT & Claude Hit Context Wall

ChatGPT and Claude both developed memory features that converged on a similar pattern before encountering limitations. Here's the thing: ChatGPT's initial memory, released in February 2024, stored user facts in a list. This meant users had to manage these memories themselves, and they could become outdated. For example, an entry like "Shlok is going to Bengaluru" became stale when he moved. What's interesting is that ChatGPT's Memory v2, in April 2025, shifted to a running profile updated in the background. This moved the management burden off the user, but the issue of staleness persisted, leading to fabricated memories. Claude approached memory differently. Its first version in August 2025 used on-demand retrieval, searching past conversations only when needed. However, by September 2025, Claude also introduced a running profile, though with a different implementation, including a visible raw profile and more frequent updates. The bottom line is that both systems faced challenges with memory management and keeping information current, highlighting the complexities of creating truly adaptive AI.