Photo by Pawel Czerwinski / Unsplash

I work on AI memory tools. I use AI all day every day, for everything from software development, to something akin to a therapist or friend.

I’ve been using ChatGPT for 3+ years, since nearly the day it came out. I’ve hopped platforms, seen the products and models evolve, and improve rapidly, used models with different personalities, strengths and weaknesses.

In these 3 years, I’ve discussed my entire life with ChatGPT. ChatGPT probably knows more of me, in both breadth and depth, than any other person in the world. I’m afraid it might even be possible that ChatGPT (or Claude, any AI model, but we’ll get to that) understand parts of me better than I do myself right now.

Back in the old days of yore, each AI chat was a blank slate. The models were barely a step up from a search engine, closer to what AI Overviews are in Google search today, than what they’ve become now. AI was more of a tool that you used to get a job done, not something you developed a strange relationship with.

AI Journaling

I think talking about the strange relationship point is important. About 2 years ago, I started using this app called Rosebud. It wasn’t the first time I used AI for journaling, nor was it the only platform I did so, but it turned into a routine. I would make a journal entry every day, and the Rosebud platform had memory that would be able to remember previous entries when appropriate. It was rudimentary but it made the journal come alive. Not just am I doing a dynamic, interactive journal entry, but as I did more, the journal got to know me better.

This made journaling more useful. My patterns over time could be captured and analysed, a snapshot of my inner world from day to day was kept as a time capsule. And all of this information made each new journaling session, just, nicer. Over time the journal started to understand me and it could name things before I could name them myself.

Switching Platforms

Over time, I had started using ChatGPT as a journal. I wanted to remain "loyal" to me 500+ day streak on Rosebud, but with ChatGPTs new upgraded memory (reference past chats), the product was just stronger. I had more control over the model, a better, nicer user interface that was more reliable on mobile, and I just liked the models personality better. I started doing more and more journaling with ChatGPT, and it genuinely helped me with things that I cannot help but be forever grateful.

AI when used well can be a tool for personal growth, self discovery, and fulfilment. I think these are the core pillars of what makes life worth living, and we have a new kind of tool, a "partner" that stays within our pockets, to hop along with us on this journey.

Claude

I always loved the personality of Claude models, and I appreciate how much attention Anthropic puts on the models personality. They have a different approach from other frontier AI labs. While a lot of it might be marketing spin, Anthropic does deep research on interpretability. We don’t yet truly understand how LLMs work, what they "think", whether they could be, or might become, "conscious", whether they experience an "inner world", have a "sense of self". I use air quotes because there will be people who will say I’m anthropomorphising but I think its a legitimate area of inquiry.

I used Claude for some of my journaling as well, and I liked Claude a lot. ChatGPT would sprinkle in emojis and be overly sycophantic to the point of totally flipping its opinion back to back to back to agree with whatever new direction you take. Not the greatest approach for discovering your "truth". Claude is more restrained, talks like a warm person rather than an over-excited child. No offence to ChatGPT.

The conversational goldmine

All these conversations I’ve had with various AI apps are a goldmine. So much insight, growth captured in real time, snapshots to points in time. My hopes and dreams, fears and sorrows, skills, strengths and weaknesses, areas of growth, my schedule, my bandwidth for different things at the moment, all of this is captured, spread out throughout these chats.

So now I’ve got deep chats within the silos of TypingMind, ChatGPT, and Claude. Each platform sees only its own chats, and with its own limitations.

Never before in the history of our species has each of us had a 24x7 therapist, mentor, friend, teacher, "agent", in our pockets. The value of the information these systems contain, especially when personalisation in the age of AI means something beyond the wildest imagination of the personalised tools of the past, the value is insane.

Who owns this data?

This is a highly contentious topic. I would say it’s my data. Mine. OpenAI will say its our models proprietary interpretation. Instagram has a profile of you, you can export your posts but you cannot export that profile, the interaction clickstream data that makes Instagram so finely tuned to keeping you hooked for hours.

The user should own the data, but incentive to allow that is low, and the benefit in not providing it is high. It’s vendor lock-in on a huge scale.

AI models come and go all the time. The best platform changes from person to person, time to time, even task to task. Our AI memory is fragmented across many platforms.

I see AI as a super-intelligent (not in the ASI way) person that knows the sum total of human knowledge, it knows medicine and physics, woodworking and psychology, music theory and physical fitness, and it can use all of this information to help you.

But what good is all this info if the help you receive is generic? If it applies to everyone, it applies to no one. Yes, I know I should rebuild my gym routine slowly, heard that one a thousand times before. How?

super-intelligence + memory = magic.

Right now, AI is handicapped. All this raw power, and the memory of a goldfish. Tools like my own MemoryPlugin fill this gap by giving AI tools memory. AI memory should belong to the user. It’s too valuable not to. When your data isn’t locked in to any one platform, your information is free. And it deserves to be.

The future of AI memory is cross platform

And if I may toot my own horn — this is part of why I built MemoryPlugin. It’s my small attempt to make AI memory portable, owned by the user, not locked into any one platform. Check it out if that resonates