It's the end of 2025, and a generally crazy time in the AI boom. I don't suspect things will get less crazy any time soon. I think there is too much big hype and big fear going on, which is valid, but I want to talk about the small, specific things that I'm optimistic about in an AI future, to look back on in 5 years and see if they came true.

Single, Unified personal data feed.

I want all my personal data across apps and devices to be fed to one centralized location (ideally a server that I own), so that I can design (with AI-assisted software development) cool dashboard and displays to analyze my habits and behaviors so that I can optimize how I live my life holistically.

For example: I listen to music on Spotify and Google Music. I read and like posts on Twitter. I use a habit tracker called "Habits" on Android. I wear a Fitbit 75% of the time, with vitals tracked there but also in Samsung Health (because I have my phone on me 100% of the time). My Google Maps has the timeline feature enabled, so I can see my location going back more than a decade. I use Strava for tracking runs. I self-host a personal Zulip server for taking notes to myself.

Many of these apps actually could import and export data from each other, but it's a pain. AI can make this easy for me in the future, and as the marginal cost of software goes to zero, I hope to be able to clone my own versions of everything, and have my own AI-powered software just ingest as much raw data as possible.

Will this ever exist in a consumer version that is good about consumer data privacy and model ownership for people who aren't nerds like me? I'm not sure, but this brings me to my next point.

AI that prompts me

ChatGPT's Pulse feature is quite interesting, but it has a lot to improve on. I'm excited for this to be a big part of the future, rather than me prompting the AI every day, it prompts me when I wake up and at various points, based on what I'm up to, based on the long term goals that I set and what I want.

Currently, there are two issues with AI prompting me:

  1. It doesn't have access to enough information
  2. It's not quite calibrated to do this well

Consider a personal trainer: they have been calibrated over their career on how to help someone in the gym, and they can watch your movements and improve your form. A future AI, with just access to your phone's cameras to watch your motions and an audio feed into your earbuds could do this, but nothing is quite at that point yet.

I also don't think that a "god in a datacenter" model will ever be calibrated enough to do this in one API request, and then go solve a programming challenge in the next API request. We need continual learning and a way to fine tune a set of simpler weights to calibrate a model for helping humans like this. This feels imminently tractable in the next few years.

Teaching

I strongly agree with Karpathy's thoughts on this that he shared on the Dwarkesh podcast. I want to be able to effortlessly learn things. Current AI's are already very helpful in learning, but there is SO much potential for improvement. Current problems:

  1. The model doesn't have a good way to really know where you are in the learning process, as Karpathy articulates. This means that it doesn't know the exact right difficulty quiz to apply.
  2. The models can only quickly generate audio and text. Complex things, especially in math and physics, are best learned with diagrams. Nano Banana Pro has made huge strides with good diagram generation, but it's still lacking the needed quality for teaching, and it's too slow.
  3. There are not good surfaces for the model to prompt me and remind me to study. Following from the above section, we need good primitives for the model to inject a thought into my mind at the right time to get me interested in continuing to learn on a subject.

When AI improves at this, learning will likely get very addicting to the right kind of person. It already kind of is.

Improved visual generation

Following from point 2 above, I'm generally just excited about better visuals. Except, I'm really not excited at all about video, I think that is too flashy and gimmicky, and less practically useful.

I just want two things:

  1. lightning fast, high quality image generation, with the ability to accept images as input for consistent style.
  2. Good diagrams, with the ability to iterate on variants of the same diagram.

Each year, I see good progress on both, and I'm excited for that to continue.

Bringing fiction books to life

I've always had a hard time getting into fiction books. Every now and then, I'm able to and I love it, but most of the time I can't get hooked. I think this is because I've always been somewhat bad at reading language that describes a scene, and then visualizing it.

I prefer the writing in books when it's only writing, rather than something like manga. I think the restriction to words is good for the quality of the storytelling and the artistry of the words themselves, but having additional visual aids on the side for a text-only fiction story I think would be the perfect combination.

As a kid, I really liked the Leviathan trilogy by Scott Westerfeld, because it was like this. The art was drawn by Keith Thompson. But, not all books can be like this, and here is where AI can step in.

This is 100% possible with today's models, it just needs a good interface. Maybe I'll build this.