The rush to become an interface to "AI"

AI is the current thing. While there are still some core questions about it up in the air, the industry is swiftly navigating itself to leverage this platform shift.

This swiftness is a particular trait of the Valley. People went all-in on almost everything since the web: mobile, Cloud, SaaS, crypto, and now AI. It's both incredible and terrifying. Of course it's also ironic, given "the current thing."

Aside: "AI" is a stand-in for a bunch of technologies. In 2024, it stands for Large Language Models (LLMs). At its core, these are RNNs but "trained" on incredibly large data sets. Think, a good majority of the words on the Internet.

Chatbots are the primary ways in which users interact with LLMs today. Yet, all of the current LLM developers - OpenAI, Google, Anthropic, Mistral all monetize their LLMs primarily via API access. They all differ whom they charge API money for and how much they charge for what. This has spurred a race to build interfaces to the LLM.

I love interfaces. And I personally believe that the chat bot and the text box will not win out in the end. I am not the only one[12.

These are things that most of the tech world has already learned: Google - famous for its simple text box - learned that English is about the only language that's easy to type text in; Google search has a very browsey interface for mobile in the emerging economies; Whatsapp's audio messages are incredibly popular in Middle East and Northern Africa and even in some European countries compared to text messaging.

Aside: Did you know that horizontal carousels perform significantly worse in mobile first areas of the world compared to desktop first areas of the world?

As someone that loves interfaces, I started collating a list of various attempts at building interfaces to LLMs. My goal is to try and keep this list updated to the best of my knowledge. If you know of something that I missed and believe it should be included, leave a comment or send me a note: gurupanguji@gurupanguji.com

Chatbots

Most people know of LLMs via chatbots since ChatGPT's introduction.

• Gemini
• ChatGPT
• Microsoft Copilot
• Claude
• Meta AI - in Whatsapp, Messenger and Instagram

Search

Yet, most people likely encounter them either via

• Google Search Generative Experience
• Bing: is increasingly pushing users to Microsoft Copilot

Productivity apps

• Microsoft 365
• Google Workspace
• Adobe GenAI suite, also for creative apps

Creative apps

Image

• Midjourney
• Stable Diffusion
• Adobe Firefly

Audio

• Suno: Generate music from text prompts
• Udio: Generate music from text prompts

Video

As far as I know, text to video generation does not have specific applications .. yet. Here are the models that are picking up steam in the world.

• Sora
• Google Lumiere
• Stable Video diffusion

Hardware

Dedicated AI hardware gadgets

• Rabbit R1
• Humane AI Pin
• Meta Rayban glasses
• LG AI mouse

Operating Systems

• Android: Gemini is currently available as an "Assistant" replacement. Predicted to become the default Assistant in Android 15 by IO 2024
• iOS: TBD
• Windows: Copilot
• Mac: TBD
• Chrome OS: TBD
• Linux: TBD

Browsers

• Chrome: Chrome has AI experiments that you can toggle - Smart Tab Organization; Generate your theme; Generate Text for any Text box
• Safari: TBD until WWDC 2024
• Arc: Max features and Browse for me via Arc Search
• SigmaOS browser

• Opera: Aria - Opera's AI assistant
• Brave: Leo - Brave's AI assistant
• Vivaldi: Nothing yet and don't see any specific plans
• Orion: Nothing yet but they are fundamentally tied to Kagi - so still holding tight to determine what their plans could be.

Developer interfaces

• Raycast: Raycast is spotlight++ on Mac OS. It has a cult following and is prescient in its support for LLM connections. They provide a Pro subscription with their own integration with multiple LLMs. They also allow individual OpenAI, Gemini, Claude extensions to bring your own API key to have an always available chat interface just a command+space away.
• Alfred: Alred predates Raycast. However, since Raycast's found an audience interested in connecting with these transformers, Alfred's building its own flows to connect to LLMs.
• llm: CLI tool by Simon to setup your own LLM in your terminal primarily focused on OpenAI models, but can be setup with other models with some tinkering

_todo aka coming soon:

• Thoughts on how each company / ecosystem is poised to deliver
• Where's the action going to be: model / apps
• Will there be a fundamentally new interface or are LLMs going to take over existing interfaces? If so, which ones?