Quick notes on AI Personal Assistants / Second Brain

Surprise surprise - I've been having multiple conversations with people on the AI Assistant / Second Brain (or whatever you call it) themes.

I thought I'd pen down my thoughts (to further my own clarity AND get inputs and feedback).

What's Inevitable:

AI enabled assistants and second brains will be a way of life - as common as having a phone is, today. Within the space - people will have their preferences, one's likely to see niches because of specific use (academia, athletes, healthcare), form factors of hardware (lesser variance here for mass usage, but specific solutions for niches), positioning of software (privacy first, secure, cheap) - quite like we do in smartphones.

To me - this isn't up for debate - the timelines in which it happens, regulations around it, what not - all of it yes, but directionally - this is where we are moving to.

That said - if anyone has a strong counter opinion - please tell me what you think and why.

Hardware

In my (possible naive and thus wrong) opinion, a lot of the hardware needs for an AI Assistant aren't too complex - and are largely solved problems in context of consumer hardware.

Audio in / out. Battery life. Vision in (out in due course as we start to have AR interfaces more commonly - but that might be a while away). Storage. Water-proofness. Compactness. Durability. These are hard engineering problems - but fairly solved ones by any stretch. Even today we have a whole range of wristbands/smart watches, phones and what not - on all spectrums of ability but with most of this basic stuff baked in.

If the hardware components are more commodity - could the form factor be the vector to innovate?

Yes and No.

I have found it hard to be convinced that anything but the Ray Ban Meta Glass equivalent is the form factor of the future. I picked them up just over a month ago, and have used them a fair bit since. Clearly, I loved them.

Why am I so bullish on glasses?

• With increasing screen time in our lives, a lot more of us are myopic - leading to a larger % of population needing glasses anyway.
• As screen time increases (and glass companies seek to make $), consumer education on using tinted no power glasses to minimise blue light impact will increase - leading to them becoming common even for the few blessed non myopic peeps.
• The form factor + spatial position of glasses is great. It allows for multi directional audio input with specificity to ensure sharpness. Single direction audio out (or multi if you want), which ensures a degree of privacy of my AIs responses to me. Body has enough room on it to bake in a battery everywhere possible. The case is easily replicable as a charging case. As and when we have vision out become a thing - could do that well. In heaviest of rains - our faces are what we protect the most so likely to have to deal with minimal water exposure (vs rings / watches / ..).

This said - I do think there is room to innovate on form factors for niche use cases with specifcic needs and constraints - where even though markets might be smaller, picking and owning them could be a significant prize. (bad analogy but how scrubs have a unique set of fabric + cut needs for medical staff).

Software

The first and fairly critical step to AI assistants (and other applications) is being really accurate on speech to text, with voice recognition (and memory) built in.

This is however something a lot of other AI applications need as well - all the AI note takers, AI support agents et al, for instance. (shoutout to Varadh Jain who's building Broadcast), and there are lots of folks pushing the boundary on this; with some choosing to build, and some to use available options.

I'm less sure this is where the value creation piece is, for AI Assistants.

Knowledge is Power. And that's where I feel most value creation will lie.

In capturing and organising knowledge, making sense of it, and crossing over to taking actions basis it. The steps below are my simplistic take on what's involved.

The second step is to store the text - a knowledge graph is the obvious way to go here. People being one set of nodes is likely (and something I'll come back to) - but this step is critical. Without proper conversion of all the info and text to graphs, with the right clustering - the 'brain' is unlikely to function. The Roam and Obsidian users have been rawdogging this for a while - but that broad structure is still the best way to organise and access information we gather.

Thereafter comes retrieval - how to surface up the right responses for questions I ask my AI Assistant, with no halucations, and consistent + reliable responses. Moving from probabilistic to deterministic is imperative (and shoutout here to Vivek Khandelwal and CogniSwitch Inc. ) for this to work.

Moving from here - is where I see a lot of opportunity, specific to AI Assistants and Second Brain products.

A framework for users to allow their AI Assistants to share (boundaried) information with each other - to allow more meaning to come out. For eg. My assistant could figure out that a portfolio company founders assistant recorded that they are looking for , (even if that's not a conversation I had with them) and nudge me that my knowledge graph suggests I can or know someone who can help them with that. Or for information shared within family members, that help be more empathetic.

What we can get assistants to do once they have all this information and context at hand - is what can start to become magical.

Choices in Building

I strongly believe that a young company should be focusing on something specific in all I laid out above. With a rapidly moving space and industry, you've got to put most of your energy on a single vector to be and stay at the cutting edge.

Further, I feel that hardware as a vector is not the best space for companies to focus on innovating. Mostly because in a mass market segment - there just may not be that much to innovate which means products will be fairly easy to mimic - and it could easily end up a fairly crowded space, with multiple players and similar looking hardware.

Though - there's definitely opportunity in building more specific hardware, for niche use cases.

The software (and more specifically the knowledge layers) are where I feel huge value can be created.

I would love for companies to focus on that, using whatever basic hardware they can develop, or other hardware they can integrate or partner with, and push boundaries to get to rapid functionality on the most critical and needle moving aspects of having AI Assistants adding leverage to our lives.

Please give this one some extra love. I had completed this post, before I lost it to LinkedIn acting up and not having saved a draft; and then nearly didn't write it again but well.