The Dream AI Hardware

I wrote a book on AI development, exploring how it works, its history, and its future. Order it here, ideally in triplicate!

---

Happy Friday! This week, first a guest post by Jason Levin, who writes Cyber Patterns, on a fascinating topic, the rise of new AI enabled hardware we can all buy and use. I’ve been thinking about this for a bit, mixing a feeling of discomfort and excitement. My thoughts after the article.

---

My #1 KPI in life is serendipity. In fact, I try to optimize my life for serendipity.

That’s why I pre-ordered Rewind’s Pendant this week, a piece of AI hardware you wear around your neck which transcribes all your conversations. 

Some people think that wearing one of these AI devices is like wearing a wire and being a Fed, but I say you’re optimizing for serendipity. Let me explain.

There is nowhere filled with more serendipity than my home of New York City. 

In the span of two hours, you can run into a high school friend you haven't seen in a decade, see a billboard with copywriting so clever you’d think it was written by Ricky Gervais, and have a wonderful conversation about everything from artificial intelligence to pizza while walking around Central Park. There’s so many opportunities for serendipity that it’s hard to keep track.

The recently released AI hardware—Rewind’s Pendant, Avi Schiffman‘s Tab, and Humane’s AI pin—promise to help you keep track of that serendipity. As a writer who makes his living on the internet, here’s how my purchasing thought process went: the more serendipity, the more ideas—the more ideas, the more publishing—the more publishing, the more views—the more views, the more money. $59 for more ideas? That’s a no-brainer investment. I already pay $12.99/month for Readwise to remember what I read and $10/month for Otter AI to transcribe my voice memos. What’s another $59? For a device to remember all my conversations? Am I being the midwit in this meme right now? Perhaps, but I don’t think so. I think I’m just a normal dude trying to capture the serendipity of life.

It’s my hope that whenever I have a cool conversation with a friend, I’ll be able to rewind word-for-word whatever was said. In writing, it’s a common idea you should write how you talk; so if I could get a recording of how I said an idea in a conversation to a friend, I can just slap that into a blog or a tweet. Or I can quote a friend in an article if they’re cool with it. Or maybe I’ll use it for those networking events where you meet 50 people with similar sounding jobs so I can remember who the hell each person is. A man can dream.

And dream I will—Here is my dream user experience for AI hardware. Rewind’s founder Dan Siroker and I follow each other on Twitter, so it’s my hope he sees this and is like “wow this is a good idea, let me implement it immediately (and pay Jason and Rohit in stock shares!)” 

Dream UX of AI Hardware: 

Exact Transcript: I want the ability to get an exact time-stamped transcript of the day’s conversations. This is where Avi’s Tab and the Rewind Pendant differ. Tab uses AI to summarize what you’re talking about but has no transcripts, whereas Rewind’s will have full transcripts. I want the full transcripts because ideas are often misinterpreted by an ultra-literal AI (think metaphors, nicknames, slang, etc.). Give me the exact transcripts or give me nothing. 

Searchability: Ability to search through transcript with a simple command-F query and copy-paste text easily. I can imagine myself thinking “I remember we talked about X → search X”. I’d also love if the device could capture locations so I can do something like “I had a good conversation in Central Park but I forget the topic → search Central Park”. 

Design: Both Tab and Rewind are going with the necklace route. I’m a fan of it. I think it looks slick and you forget about it quickly (if it’s lightweight). The recording/not-recording signal needs to be obvious like Snapchat’s Ray-Ban glasses which glow red when on—otherwise anyone wearing it turns into a Fed and we are officially bringing back sauna meetings. 

Daily Wrap-Up Emails: Give me the day’s summary of my transcript with the option to view the whole thing in-browser or in-app. This is the Readwise daily email model. 

AI-Suggestions: OK this would make it an absolute 11 out of 10 but what if AI hardware could give you further things to research based on your conversations? So let’s say I talked about JFK for 10 minutes in a conversation with a friend, the AI knows, and suggests books and documentaries for me to look into. Maybe there can even be a Creator Option where the AI suggests content ideas based on conversations. Imagine a notification “Turn this into a tweet”. 

P.S. Hey Dan, I hope you read that part about giving me and Rohit stock! 

P.S.S. If you need a new name for the Rewind Pendant, may I suggest The Serendipity Machine?

-- 

For more of Jason’s writing, subscribe to his weekly newsletter Cyber Patterns. You can also find him tweeting thoughts, memes, and bangers @iamjasonlevin.

2/ More thoughts about AI hardware, and making glasses cool

The key question I wonder about AI hardware is whether it’s an inevitable consequence of technology marching into the frontier, or if it’s a fad that we won’t remember in a few years. And the answer here is probably going to come from the question of “why”, why someone would choose to use one of these things.

There’s a thing we often say, of how we walk around with a tool for mass surveillance in our pockets, willingly. It works though because it us useful. It combined pretty much every previous technology that we used into one small package.

The AI enabled accessory I’m most excited about is, to my utter surprise, Meta’s Rayban glasses. The main use case is not having to pull out my phone to take pictures of my kids.

---

3/ It’s also how OpenAI plans on creating their own hardware project

Along with trying to think about how to create their own chips, OpenAI is also realising that a chat window might not be enough of a draw. Following in the footsteps of Meta and Google and Microsoft and Amazon and pretty much every software manufacturer who wanted to make hardware, now it’s OpenAI’s turn.

What this might actually be is as yet unknown. I wonder if it’s going to be a redo of creating another phone (likely to fail) vs creating something like a competitor to Rewind (also hard, because what’s the point) vs creating an Alexa competitor (likely to work!).

Though might want to hurry, Google are already going to add it to Home and I’m sure the rest aren’t far behind.

---

4/ Toys!

The thing that I haven’t yet seen, but can’t be too far behind, is AI enabled toys for kids. One of the more annoying things as a parent is the proliferation of toys that talk, hoot, snort, growl, yelp and honk at the slightest provocation, around your house. I sort of want the ideal toy as something which the child can talk to and interact with, and ideally be the sort of companion that can help her learn.

But of course it’s complicated for many reasons. Considering the direct link to the internet and the very likely possibility of prompt hacking, this feels like a difficult, possibly impossible, line to tread.

On the other hand, how many kids have you seen with their parents mobiles? Or iPads? It feels inevitable that this happens. Our stories consistently have talking animals and companions, makes sense we’d make them true. My biggest question is whether this is likely to be a device specific innovation or software that makes it easier to integrate things faster.

Plus, the idea of making one of these properly would have to include adding memory to them, which would need some fun engineering, and will have pretty great spillover effects. In fact, I think Rewind et al will probably do an API soon too soon enough.

---

5/ AI powered drone beats human champion pilots

The future is here.

The latest mortals to feel the sting of AI-induced defeat are three expert drone racers who were beaten by an algorithm that learned to fly a drone around a 3D race course at breakneck speeds without crashing. Or at least not crashing too often.

A weird part of racing anything is that it only really makes sense from a human perspective. We have maximum g forces we can endure, and reaction times that are hardcoded by biology, and neither are particularly optimal for racing around a 3D arena (or a 2D arena), compared to perfectly tuned machines.

So I’m not surprised this happend. Drones are already doing jobs that humans find taxing. Military reconnaissance is the easy one, but did you know we have drones to count sheep? Presumably because the humans who used to do it fell asleep the drones had a comparative advantage. They can also count whales and help 3D model buildings.

Anyway, as the owner of a new toy drone, these things are remarkably cool, and we’re most likely going to live in a world ten years from now where the skies are much more crowded than today. Or so we should hope.

---

I do think the near future will include many more hardware devices that allow us to interact, even if via chat, and having the ability to interact with them in ambient mode will likely make the world way more exciting.