Mem is a leader in the crowded application category for personal productivity note-taking and task wrangling or “second brain” applications. Mem is well funded1 Series A funding at $29.1M, OpenAI contributing $23.5M round in Mem, an AI-powered note-taking app, from TechCrunch. and has a passionate user community on Slack. I am a Mem subscriber.
I like Mem because the interface is simple, it implements a reasonable subset of Markdown2 About 60% of the Markdown Cheatsheet is implemented, but no tables yet. Github wiki., has built-in templates to streamline your work, and Zapier3 Available from Zapier integration studio: Zapier Studio. integration to load content into Mem. Mem continues to iterate the product, but in what direction?
Chat is Mem’s AI implementation4 Previously called Smart Text or MemX. During the beta of MemX (now Chat) I observed the Mem syntax was very “technical” and could benefit from clear sentence structure in the prompts. See MemX is too Opaque. to surface Mem content or support writing. The design team continues to experiment with large language models (LLMs) and Chat user interactions. My conclusion is Chat does a good job summarizing meeting transcripts or re-drafting communications. The challenge with Mem Chat is an interaction model that leads to multiple dead-ends.
As everyone explores prompt interfaces or new AI-assisted interaction models, what is ideal? How much should the application reveal about decisions underlying the suggestions?
Overall, the Mem design team has avoided Norman’s criticism5 A recent paper from the astute Neilsen Norman reviews the new AI interfaces. He complains there is too much effort demanded by the user to construct useful prompts. AI: First New UI Paradigm in 60 Years. that AI chat interfaces are the “first new UI paradigm in 60 Years” where users tell the computer what to do. Unlike the prompt engineering of ChatGPT where you describe the desired context and outcome, Mem tries to determine your intent from the content of the Mem and associated Mem collections.
I will defer the rise of the ✨ spark icon as the choice to represent AI assistants in products such as Google Photos and Notion AI. Some say “it is widely recognized as a symbol of innovation, creativity, and inspiration in the tech industry, particularly in the field of AI.”6 A positive comment on the spark icon from Bootcamp.
Many think the spark conveys the wonder and possibility of AI. However, wonder without transparency is confusion. I think Fast Company’s three icon recommendation is better.7 FastCompany AI icons for all generative text. This icon set provides for a generalized approach for all content types, hybrid content generated by humans but augmented by AI. This middle content type “AI-H” is the domain of Mem.
Mem frequently uses a sidebar to show linked Mems. The sidebar is a good choice to view content side-by-side. Mem Chat leverages the sidebar to provide suggestions. The use of the sidebar is nice; the refresh icon is nicely placed. I assume the ✨ spark icon is meant to indicate the top two sidebar elements are machine generated. Good so far. And Chat has not halucenated or produced unreasonble suggestions.
At a macro-level, there is room for improvement regarding Chat Suggestions. If I was looking for a specific answer, I would use Search. When using Chat, I am looking for suggestions or ideas framed in the context of “what sort of answers would someone like me produce to questions (or Mems) that look like this?”
At a micro-level, there are other challenges,8 Sample Mem sidebar extension with Chat. I appreciate the symmetry of three options in each box. But I need context for why “these three” are listed in either the Chat Suggestions or the Similar to this mem groups. including:
The first chat recommendation is “Provide a critique on the article.” What is the logic for using “on the article”? Was it because the Mem template is “Article” or the Mem is tagged “#article”?
When I asked Mem to regenerate the Chat, the suggestions moved away from the article theme. Chat now seems to focus on ways to improve the research.
Turning to the companion boxes in the UI, I understand the “Similar to this mem.”9 I do wonder why Mem is not sentence case. Mem is a product; a capital M is appropriate. The last box is called “articles,” which I assume comes from the “#article” collection? I would recommend a hover tip or some visual clue why these Mems and collections where recommended.
I do appreciate the ability to refresh Chat Suggestions; performance is super fast, so I assume Mem has cached suggestions. After refreshing a few times, I found a suggestion I wanted to explore.
Each suggestion is a hyperlink to the “Chat” interface of Mem. This interface is a full page with a Chat icon highlighted at the top of the page. The response is polite and useful, as follows:
Certainly, Peter. While using ChatGPT and PlantUML for mind mapping can be a powerful combination, there are a few potential limitations and drawbacks to consider:
Limited Contextual Understanding: ChatGPT, while impressive, may not always fully grasp the context or nuances of a conversation. This could lead to inaccuracies or misunderstandings when generating mind maps based on user prompts.
Lack of Real-Time Interaction: ChatGPT doesn’t support real-time interaction. If you want to modify or expand a part of the mind map while it’s being created, you might need to start over or manually edit the generated content.
Formatting Constraints: PlantUML, while versatile, has its own syntax and formatting rules. This might limit the complexity or visual appeal of the mind maps you can create. For instance, it might not support certain types of graphical customization that other mind mapping tools offer.
Mem presents two options for the Chat Suggestions: save as a new Mem or Copy the text. These are nice, productive options. However, I frequently want to return back to the Mem that initiated this AI diversion. Does Mem provide any navigational aids next to the Mem or Copy options? No.
Mem does provide a disclaimer below the text “Mem answers questions and creates content using the knowledge you’ve saved and public knowledge,” with a nice learn more link.10 The link is to Mem support article: Chat with your Mem. Mem, however, gives me no UI element return to my work. The browser back button or Alt+Left Arrow is available. I would expect the UI to help with the following:
Gracefully return back to the Mem that initiated the Chat suggestions.
If I opt to create a new Mem from the generated text, Mem should show the original Mem and the new Mem in the split-pane interface.
Place the thumbs-up, thumbs-down rating of the generated text in the same UI location as the new Mem and Copy options. Instead the rating buttons are at the top of the text; the Mem buttons are at the bottom.
If the goal of Mem Chat is to improve my productivity, then don’t erase that productivity with splintered menus and awkward navigation. My recommendation is to make the Chat button in the top navigation a toggle. The button is active when in Chat; clicking the active button will turn Chat off and return to the original Mem.
I understand the Mem design team wants to retain color in the top menu. A possible menu redesign would use color to show feature state (on is color, off is white) and to simplify the menu as shown below.
These small interaction changes would reduce user fatigue and navigation challenges; notably there is no way to navigate “home” without sliding out the left navigation, which most Mem users keep closed to maximize screen real estate.
Imagine you are a party with friends and new acquaintances. You can pick between two questions: “what are your goals?” or “are you a coffee or tea fan?” Perhaps the reaction is different based on the relationship with the questioner. Nearly everyone would perfer the the second question because it is easier to answer; it gets the conversation started and might lead to bigger questions.
Mem starts with a Chat question: “What can I help with?”
While a helpful question, I have yet to get a satisfactory response from Mem. Most days I respond “what are my tasks today,” and Chat responds “I don’t have real-time access to your task list or calendar.” The Mem Slack comminty helpfull notes that LLMs “don’t have a sense of time, things like today, tomorrow.” However, Mem does have a Markdown convention of “[]” to create a task; Mem also has collections to group Mems around topics. Mem even shows “consulted sources” when generating suggestions. Finally, Mem suggests a prompt “What tasks do I have for tomorrow?” that produces the same “I don’t have real-time access” response.
My recommendation is don’t present time-based prompts that Mem Chat cannot solve. Like Mem navigation, these dead-ends create cognitive frustration.
Returning to example scenario from the party, what if Mem started with these prompts:
Mems from yesterday focused on these topics [list the top 3 like Mem did in the Sidebar example above]. Then Mem asked “What topic should we continue with today?”
Even better would be:
You have the following meetings this week [list]. Do you need help drafting meeting invitations or review the email before you send it?
Generative AI is a journey. I appreciate the partnership with the Mem team on the journey. I’m prepared for the journey, I just want the destination to emerge. The more Mem can help, the more productive we can become.