AI Memory Lock In Effects

brown building near body of water under blue sky and white clouds
Photo by Mike van Schoonderwalt on

This morning I was helping a portfolio company write an update that needed to go out to investors, partners, and prospective customers. In it the importance of being brief, clear, and direct were paramount. I found myself using my usual toolbox to improve things which was a combination of talking it out with a voice to text capture, condensing things into a few key bullets, and finally “jamming” with ChatGPT to further refine my thoughts. I am a very collaborative writer and builder. I benefit from talking things through with someone. Since that is not always possible, the advent of AI has been extremely helpful.

Today, I realized that ChatGPT now has memories across posts.

In April 2024 ChatGPT has this to say:

  1. Persistent Memory: When you tell me something you’d like me to remember, I can store this information and recall it in future conversations. This helps in maintaining context and continuity, making our interactions more personalized and efficient.
  2. Privacy and Control: You have control over this memory. You can ask me to forget certain things I’ve remembered or to review what I currently remember.
  3. Dynamic Updating: As our conversations evolve, I can update or modify the memory based on new information you provide. This ensures that the memory remains accurate and relevant to your needs.
  4. Contextual Usage: The information I remember is used to provide more tailored responses and suggestions based on your specific context and preferences.

I have been thinking about this concept for awhile and 2018 I described it Algorithmic Lock In as it related to Spotify. the more you use it, the better it becomes. Its very hard to migrate your taste profile once it has been established. The other is the Data Exhaust that similarly comes out of a service you use. My articulation of this from 2008 was “The more data left behind, that can be harvested, the better the underlying network.” This is not to say that the data left behind can’t be harvested for evil instead of good, it just means that the lock in effect is real.

This brings us to my point today, which is that ChatGPT is now building a user data moat. The more you use the OpenAI platform, the more data it ingests about you. You can control, edit, omit, and contour this information (as described above) as you see fit. This is a fascinating new way to think about the moat of a AI business. As AI companies are struggling to support the huge compute power necessary to run against smaller revenues than expected, the more “sticky” the platform the better chance they have of succeeding. In this NYTimes article it is described that startups in the space have little chance to succeed without the financial support and partnership of a large platform in the space – enter Microsoft, Meta, and others getting into the game. This David+Goliath match makes for a very deep pocketed, and now potentially deeper moat standard.

As ChatGPT continues to build and iterate, there is a high likelihood that my past actions and energy put into it will benefit me more – thus justifying the subscription even further. AI is getting smarter everyday, and the AI you use everyday is getting smarter for you to use.

Previous articleBurnout vs Burnin