ChatGPT for Memories: What Happens When GPT-4 is Trained On You
ChatGPT and its endless spinoffs are augmenting the brain in every which way. And now, there’s even a GPT-4 application for augmenting human memory.
WTF? ChatGPT for ‘ME’mory
Rewind believes that LLMs aren’t just useful for generating net-new information based on their vast knowledge set, but can also be used to recall information we’ve interacted with. They basically ask the question: What if GPT-4 were fine-tuned on only YOUR data to learn who you are and what you consume, create, and communicate?
ChatGPT for Me is what they’re calling this AI-enhanced memory application. It’s a co-pilot for the mind that captures everything you’ve seen, said, or heard online – whether through emails, messages, web visits, content consumed, or meetings attended. The ChatGPT-like interface then allows users to ask the AI to help remember anything. The promise is to give users a second brain with perfect memory.
Some of the prompts they showcase ChatGPT for Me being able to act on include:
- Interactions – Determine how you know someone, what you last said to them, or when you first met them.
- Time – Figure out what you did last Friday or ate on Tuesday.
- Learnings – Find the article, video, or meeting where something along the lines of X was said or shared.
- Browsings – Repopulate the products you looked at purchasing and connect dots across multiple sources of information.
If it sounds invasive to give this level of unfettered access to every action you take on your device, and if it sounds like you might end up with inaccurate and hallucinated answers from the AI, then here’s Rewind’s response:
What data leaves my machine with ChatGPT for Me?
A search is done locally to find the most relevant moments based on your question. Rewind sends only the text from those moments to OpenAI in order to generate an answer to your question. No video recordings or audio files ever leave your machine.
How can I know the answer provided by ChatGPT for Me is correct?
Every ChatGPT for Me answer includes direct links to the very moments in time that shaped its response. Because we always show you sources with every response, you don't need to worry about any hallucinations or any inaccurate responses.
This is certainly one of the more “WTF” use cases for GPT-4 that I’ve heard of yet. I’m not sure if I’m sold on the idea of bringing AI that close to my workflow. A part of me is still trying to be cautious and careful with what data I willfully give to these LLMs.
But this does support my theory on chat conversations being the new Internet cookie (Everydays 192) and GPT apps being used for data farming (Everydays 194). If Rewind can get it right with ChatGPT for Me, they’ve essentially created “Atomic Habits Part 2: Habits Mean Nothing in the Age of AI.”
On an unrelated note, ChatGPT has joined America in throwing shade at the state of Ohio:
Member discussion