I have a question about the optional “Import Memory” connection for ChatGPT and Claude and the proposed prompt that accompanies it in the app.
I understand the goal of bringing context from another AI system, but the suggested prompt seems much broader than what a note-transcription application appears to need. It asks for things like behavioral patterns, psychological insights, and a comprehensive “Behavioral Profile” including “Recurring anxieties, emotional triggers, likes/dislikes, and subjects of passionate curiosity or aversion.”
Context that improves transcription and summaries—names, recurring terminology, communication preferences, and perhaps current areas of focus make sense. But I’m curious why the suggested prompt goes well beyond that?
I’ve also noticed Pocket appears to learn over time by incorporating information from transcripts and summaries into its memory. If that’s part of the product vision, I’d love to better understand what gets promoted to long-term memory, why, and how much control users have over reviewing or editing it.
Is the long-term vision for Pocket to become something more like a true AI Chief of Staff, or is there another reason the suggested import is so comprehensive?
1 Like
I’m a bit behind on the posts these days because I am neck deep in AI Theology. And as such, I too, would really like to hear from the guys on what the thought process is behind such a broad prompt for long-term memory generation. I have not read the specs of the new feature either, so I need to do that also, but it does sound like a bit overkill when training Anthropic or OpenAI models to generate Mind Maps.
Thanks for your note, and congratulations on being a “Founding Member” of this community. I appreciate folks who support young companies as they’re getting started.
Pocket is a great idea with lots of potential and smart people working on it. They’ve created an onboarding experience that’s easy and a device that’s easy to use. Kudos to them for both.
Sadly, I’ve decided that while I like the convenience of the device and its thoughtful engineering, Pocket is often unsuitable for me. Here’s why:
- The AI import prompt mentioned in my first post here is purpose built to construct a behavioral profile of the user, for reasons that remain unclear. And whatever good intent Pocket may have, this creates the impression that conversations are or may be mined for commercial purposes.
- The product builds a summary profile of the user based on the user’s conversations with other people, and there’s no way to opt out of this.
- I can’t delete or correct all information in my profile. The profile regenerates information even after I’ve deleted underlying recordings in the cloud and on device, which means some form of long-term memory exists, inaccessible to the user.
- I can’t get the AI agent to remove inaccurate or outdated information from the profile. I’ve asked the agent to do so multiple times. Each time, it said it had queued that request, but the removal never actually happened.
- The last time this failed, the agent told me to contact Pocket support so an engineer could remove content manually. Full disclosure: I did not. A product that relies on human intervention to change a profile the user never asked for, can’t opt out of, and can’t fully edit or delete isn’t well suited to sensitive discussions.
- Pocket doesn’t always match voices well, an issue I’m confident the company is working on. But because of this, “my” profile includes content other people said that isn’t true of me. Worse, it contains sensitive information of people with whom I spoke.
- Pocket’s website describes the product as an AI assistant, which is a broader role than a notetaker. If that’s the direction Pocket wants to go, it should say so plainly, explain what’s involved, and let users opt in or out.
I purchased Pocket solely for personal use, not for my job. But as a lawyer by trade, I couldn’t use Pocket for work in its current state even if I wanted to, despite the company’s Suits-inspired advertising. I couldn’t have a client’s sensitive information stored permanently in the cloud or attributed to a behavioral profile. Full disclosure: I’m using the free version of Pocket.
For now, I’ll continue to use Pocket only for low-stakes, personal purposes where these gaps matter less. I’d welcome a substantive response here from the team on the questions raised in my original post, particularly whether profile corrections and deletions are on the roadmap, and how Pocket intends to balance its “AI assistant” ambitions with meaningful user control. I remain hopeful this becomes a product I can trust with more. It has much potential.
1 Like
Interesting reading, and in some ways you’ve touched on broader AI-considerations that apply to all providers in this space.
Personally, and to rid myself of the dilemma you’ve described, I treat Pocket mainly as a go-anywhere note-taker, with the intention of then integrating those notes in another, centrally integrated and secure platform or data-base. I have not connected and exported chats from my other AI accounts to Pocket.
I take this approach for a number of reasons, including:
- Pocket is not established well enough yet for me to readily handover unlimited IP without some concern for the longevity of their service (compared to for example, Microsoft with Office365 & Copilot),
- If you’re in the Professional sector (lawyer, corporate exec, or any other), you’ll be encountering enterprise AI solutions in your daily workflow from different providers on an increasing basis (Copilot, Gemini, Claude, ChatGPT), and Pocket is not setup to be the hub but rather to be a spoke in that arrangement,
- Pocket is currently attractive because it combines an innovative and integrated hardware & software solution to anywhere/anytime note taking that currently tops other offerings - but the technology in this space is developing so rapidly, that I must be prepared to move on whenever the next-best solution arrives. So, replace the ‘spoke’ rather than the ‘hub’.
I’m sure Pocket aims to retain early adopters and customers longer by offering services that go beyond recording & transcription and to essentially ’lock you in’ to more of a system of AI administrative support - no harm or foul in that goal, it’s simply the commercial side of innovation.
Bottomline and my take - don’t miss out today on the productivity upsides of its core utility, which is anytime/anywhere recording, transcription and notes, while they figure out their proposed added-value services of tomorrow 