Exited Founder | Entrepreneur Coach | Builder

Exited founder, building #2 which will use AI to help companies be the ones that everyone in their industry chases. 

Working in Product or leading revenue? Message me about it.

Unintended Consequence: Giving LLMs perfect memory means this won’t be a one-model-wins-all market.

The upside of persistent memory (ChatGPT remembering conversations from months ago) is obvious for users (and advertisers). But the more I use it, the more I’m noticing an unintended effect:

I’m now defaulting to two LLMs for most real work.

Why? Because I need a clean slate. For example, if I ask one model, “What benefits do customers like most about us?” it will often echo an assertion I made in a previous pitch instead of something a customer specifically said. So when I want to know which benefits to emphasize in the next sales call, I’ll open a second LLM and ask it cold, with no prior influence.

Could this be solved with an “incognito mode” for LLMs? Maybe. But so far, I find myself reaching for a completely different model when I want to control for bias and accumulated memory.

Perfect recall is powerful but it also pushes power users toward running multiple models in parallel.

Patrick Randolph

2x Founder. Most recently built QueueDr to seven figures in ARR and an eight figure exit. Co-creator of The Dossier and Aimbler!

taplio

Patrick Randolph's Linkedin Analytics