Chat is overrated. Here are 4 innovative AI product interfaces

I believe that chatbots are overrated.

These days, it seems like every other AI

product defaults to a chatbot interface,

probably because Chad GBT is a chatbot,

but I think AI really needs a UI

revolution. Chat is a great interface

for some use cases, but it's not the

right one for all of them. The biggest

problem is that it places the burden of

finding use cases on the user. Its

open-endedness is both its biggest

strength and biggest limitation. It's

infinitely flexible and personalized,

but it can also be extremely confusing

for people who don't know what to ask.

Also, people don't know what they don't

know. Writing good prompt is hard. It

takes a lot of prior knowledge to be

able to ask good questions. And chat

interfaces are reactive. The user must

initiate the interaction and we humans

are lazy. Today, I'll talk about four

types of innovative interfaces I've seen

for AI products and the specific

products that inspire these

observations. The first type is what I

call async co-creation between human and

AI. The primary example is Granola, the

AI meeting notetaker. There are hundreds

of AI meeting notetakers on the market.

What sets Granola apart is the

interface. For most AI note takers, the

AI transcribes everything people said

and then you're given a summary in the

end. But Granola is different. In

Granola, you actually also need to take

notes during the meeting on a notepad

like on the left. But these can be just

rough notes, kind of like the messy

notes to sell that you type on your

Apple Notes. At the same time, the AI is

also transcribing what everyone said.

After the meeting, the AI combines your

rough notes with the transcription,

transforming your notes into a polished,

enriched, and structured version like on

the right. So, this is basically

asynchronous collaboration between you

and the AI. Because you wrote parts of

it yourself, the quality of the notes

feels a lot higher. I really like

Granola's interface because it puts

humans in the loop without compromising

the intuitiveness of the note-taking

experience. The second type of interface

is personalized content engines. The

first interface I saw in this category

was Notebook LM's audio overview and

most recently video overview. You can

upload any material like PDFs or long

YouTube videos and Notebook LM will

transform or remix it into a podcast

with two AI hosts. Their back and forth

conversation feels very natural and

makes it a lot easier to digest

difficult content. The new video

overview feature takes it even further.

It transforms your content into an

explainer video with AI generated

graphics strips and voice over.

>> Self attention lets the model for every

single word it's processing look at the

entire sentence at the same time. And

honestly, the best way to think about it

is like you're at a noisy cocktail

party. Let's say you want to understand

a specific topic. Note that this is not

summarizing content, nor is it

traditional text to speech. This is

remixing content. I like the idea of

content remix because we all learn in

different ways. Some people are visual

learners, some people prefer to listen,

others prefer to read, but only in a

certain style. With AI, we can basically

remix any type of content into any other

type of content. Say research paper to

podcast or long video to short video.

This allows people to consume content

that they otherwise would not consume.

For example, I don't have a technical

background, so I find AI research papers

very intimidating to read, but I can

totally listen to a five-minute

explainer video on the same paper

generated by Notebook LM. The AI

translates technical concepts into

layman's language, and the visuals

further help enhance my understanding.

Another product I'm watching in this

space is HUX, which is a startup founded

by the team that previously worked on

notebook LM at Google. is also centered

around the idea of the personalized

content remix except that in this case

the content comes directly from your

existing personal context like your

emails and your calendar. You can

connect Hawks to your email and calendar

and it will generate a personal audio AI

briefing with the two AI hosts about

stuff in your inbox what you can expect

for your day and also the latest news.

You can listen to it while commuting to

work. It's kind of like having a

personal chief of staff giving you a

briefing to get you ready for your day.

The product is still pretty early stage,

but given the team's track record, I'm

very excited to see where it goes and

just in general very excited about the

idea of hyperpersonalized content for

one. Another product along these lines

is Pillow Talk, an AI powered personal

voice journal app. First of all, it's

just the most beautiful AI app I've ever

used. Secondly, unlike most other AI

companions or wellness apps, it's not a

chatbot. You can yap into a microphone

about whatever is on your mind. And then

you can see an AI generated analysis

about what you just said, which surfaces

insights about your thought patterns and

characteristics. The AI is basically

like a mirror of yourself, helping you

understand yourself by seeing you from

an outsers's perspective. To me, this is

also a type of remix, turning firsterson

content into third person content,

turning the subjective into the

objective. It allows us to see ourselves

in refreshing ways. The third category

for AI interface is what I call a button

that says a thousand words. This

category consists of products that have

thoughtfully designed buttons in GUI

that convey a lot of useful information

coupled with surrounding context. For

example, I use the AI podcast app

Snipped which has a feature called

create snip. When you're listening to a

podcast and want to note something down,

you can press this button and then the

AI will take the current timestamp and

turn the corresponding transcript into a

knowledge card, like an insight or piece

of advice, which you can then save or

share. This is a great interface because

it's extremely low friction, just a

click of a button, yet it conveys a lot

of context to the AI. Like, I'm

currently at 43 minutes and 35 seconds

in this podcast episode. I find this

line particularly interesting. help me

note it down for future reference and

rewrite in a nice format. This is what I

call maximum context with minimum

effort. A similar interface can be found

in a meeting note-taking app quill.

>> This is a really cool button. This

basically takes a highlight and

highlights the thing that was just said

so that later on in the notes it will

draw attention to it and I'll show you

that later.

>> When you're in a meeting and someone

says something you want to note down,

you can just click highlight point and

Quill will note down the current

timestamp. Originally, I just wrote

this, but then what it enhanced was

these things around what when I said

them or when I highlighted them, which

is really, really cool. Isn't that

awesome?

>> After the meeting, the AI will enrich

your notes with what was said at the

specific timestamp. By clicking that

button, you're basically telling the AI

this is worth taking down. The fourth

type of interface is AI generated feeds.

I still believe the feed is one of the

best interfaces invented in the mobile

internet era because it's more

proactive. Let's face it, humans are

lazy. Most people don't want to type. We

just want to sit back and scroll

content. The point is, can AI deliver

personalized feeds that are not brain

rot that actually makes us feel good and

learn stuff. The most established AI

generated feed products I've seen are

mostly in the news category, such as

particle news. The AI remixes content

written by the media into short

articles, and you can also adjust the

tone and style. The content is delivered

in a personalized feed with

recommendation algorithm that you can

also tinker with. I think this interface

is great. However, I'm not using

particle news as much as I like simply

because I don't get my news from the

media anymore. I get my news from ads

and newsletters and other types of

social media. News is still a

productivity use case and we must

remember that most people in the world

want to kill time, not save time. What

I'm looking forward to is seeing AI

generated feeds for entertainment. We're

seeing an inkling of this in Pico and

Character AI's mobile apps. And recently

I saw a cool product called spiel work

that is basically a feed of mini games

that users have vi coded with AI that

you can scroll through and you can also

remix on top of other people's games. In

some here are some things that I think

the above products got right when it

comes to interfaces. One bring humans in

the loop at the right times in a

frictionless way like granola asking you

to take notes yourself. Two get users to

give maximum context to AI with minimum

effort like how you can take notes in

snipped with one click. Three, the best

AI products might make the AI invisible

or behind the scene. Like in Particle

News, the user only needs to interact

with the fruit of the AI's labor, not

the AI itself. Focus on the user's job

to be done. AI is just the enabling

technology that gets the job done. It's

a means to an end. Four, have the AI

come to the user, not the other way

around. Like in HU, most of the time,

you don't need to give any prompt. The

AI just comes to you with readily made

content generated from your personal

context. If we look back in history, the

personal computer revolution itself was

driven largely by interface innovations

like the graphical user interface and

the mouse. These replaced command line

interfaces and made computers accessible

to regular folks, not just scientists

and programmers. Similarly, the mobile

internet revolution was largely driven

by the touchscreen which was also a

interface innovation. Twothirds of

Americans still haven't used ChadBt. If

we really want the benefit of AI to

reach more people outside of the tech

bubble, we need more intuitive

interfaces that are truly accessible to

everyone. While we think about raising

the ceiling of intelligence, we should

also think about how to lower the

barrier for usage with innovative user

interfaces.