Leonardo.ai, Context Length, PDFs, Chat Quick-Change, and More!

The amount of significant upgrades in this latest update to Magai is staggering. Buckle your seatbelt, ’cause we’re going to blaze through them as quickly as we can.

Two of the biggest updates have been a long time coming. The ability to continue chatting past an AI model context limit was my number one priority. Closely behind that was adding a much better image generation model to make up for Dall-E 2’s lack of quality outputs.

Many of the other updates were a combination of quality of life and user-requested features that just make the Magai experience more magical.

So let’s dive into it.

Blowing Past the Context Limit

Now, most people don’t even know what a context limit is. Since the native ChatGPT app allows users to keep chatting for as long as they want without ever stopping, they just assume the AI remembers the entire conversation. After all, it’s right there on the page, right?

If they can see it, obviously the AI can see it too, right?

Wrong.

For those who don’t know, every AI model today has a limit to how much it can “remember” in a conversation. That limit is commonly referred to as either the Context Length or Context Window. The easiest way to understand it is to think of a memory stick.

You can only fit so many images on a memory stick until there’s no more room. In order to fit more images on it, you’d need to delete items from the memory and that would allow you to add more.

This is what ChatGPT and other AI Chat apps do when they allow you to keep chatting on and on past the AI’s context length. They just don’t tell you that they’re actively removing earlier message from the AI’s “memory” in order to make room for new messages.

I never liked that about AI chat apps. It doesn’t feel honest, and to some degree feels deceptive. It can also cause poor user experience when the AI produces something that seemingly ignores previous instructions or information. The user is left thinking the AI made a mistake, when in fact it just didn’t have the previous information in it’s context window (memory) anymore.

At first, I didn’t have a mechanism in place for Magai to be able to automatically remove messages from the AI model’s context length and make room for more. So users would have a hard stop when they reached the limit.

I fixed that, and built a more elegant and transparent way of removing messages from the context and informing the user.

What I’ve done in Magai is give users a subtle visual indicator of where the AI’s context is exceeded. This way you know exactly how much of the conversation the AI actually “remembers.”

Leonard.ai for 10x better AI images!

One of the most requested features since releasing Magai earlier this year was a better image generation model.

Dall-E 2 was great when it first came out, but it was quickly surpassed by other AI image models. The most popular, and by far the most advanced AI image generations come from Midjourney. And while I would love to integrate Midjourney into Magai, unfortunately they don’t have an API.

So I was forced to look around for another suitable alternative.

Then I found Leonardo.ai.

In my opinion, no other AI image generator resource comes as close to Midjourney-level creations as Leonardo does. Their system is powerful, robust, and community-driven.

If you were to get your own Leonardo account, you’d see they have a plethora of models available, as well as fine-tuning tools to help give you a massive amount of creative options. For Magai’s first implementation of Leondardo, I wanted to keeps things simple while we test it with our users.

Look for a lot of additions to come with Leondard’s model inside of Magai. In the meantime, a simple prompt is all you need to start generating way better AI images.

Chat with your PDFs!

There have been many requests from users to be able to upload a document into the chat and have the AI read it. You can now upload PDFs into your Chats and the AI will be able to read it in Magai!

The next item I’ll be working on is the ability to add .docx and other file formats as well. And if OpenAI ever releases the multi-modal ability to interpret images, I’ll be ready for that as well.

Get YouTube video transcript automatically

Magai has always been able to detect a URL inside your prompt and automatically fetch the content of the destination page. I’ve taken a step further to recognize YouTube video links and automatically fetch the transcript of the given video.

Just paste a link to a YouTube video along with your prompt, and Magai will fetch the transcript and add it to the prompt behind the scenes. Now the AI will have the transcript in its memory for you to use in the rest of the chat.

Here are some great examples of how you can use this feature in your day-to-day work:

Content Creation: Transcripts can be turned into blog posts (shortcutting the iterative approach), articles, or ebooks. This is useful for content creators to double their content output without much effort.
Social Media Posts: You can have the AI pull key quotes and statements from the transcript and use them as bite-sized content for social media posts. This can lead to more engagement and increased brand exposure.
Infographics: AI can identify key information, data, and stats in the transcript and suggest visually appealing infographics.
Interactive Content: The AI can suggest quizzes, polls, or games based on the content of the transcript. This not only makes the content more engaging but also aids better understanding and retention.
Q&A Database: Use AI to aggregate questions and answers from the transcript to create an FAQ or Q&A section on a website.
Content Translation: Use the transcript as a basis for translating the content into other languages to reach a wider audience.
E-learning Courses: The transcript can be divided into sections and chapters to create an online course on e-learning platforms.
Repurpose into Different Content Formats: Transform the transcript into slide decks, webinars, eBooks, or podcast episodes.
Text to Speech: AI can convert the transcribed text into voiceovers for animations, explainer videos, audiobooks, etc.
Chatbot Training: The transcribed conversations can be used to train chatbot systems to make them more interactive, informative, and engaging.
Visual Summaries: AI can generate captivating video summaries from lengthy transcripts.

There are way more ways you could use this feature to make your life easier. I can’t wait to hear all the ways Magai users are going to take advantage of it!

Claude v2 FTW

Anthropic recently released a huge upgrade to their AI model, Claude. Here are the main things you should know:

Here is a summary of the content:

Anthropic announces Claude 2, an improved version of their AI model Claude.
Claude 2 has better performance, longer responses up to 100k tokens, and can be accessed via API and the new claude.ai website.
Claude 2 scored higher on exams like the Bar exam and GRE compared to Claude 1.3, showing improved skills in coding, math and reasoning.
Anthropic says they are working to improve the safety of Claude 2 so it produces less harmful or offensive outputs. They use techniques like red teaming and evaluation to improve safety.

That covers the main points announced in the Claude 2 product announcement from Anthropic.

I’m already a huge fan of Claude. Not just for it’s ability to remember 75,000 words in a conversation, but for the way it accurately maintains the integrity of the content your feed it. Unlike GPT, Claude is far less likely to make “creative changes” to text you’ve given it to work with. It is much more reliable in that sense, if you’re working with pre-existing content and modifying it.

If you haven’t yet, give Claude a try inside of Magai! You might be surprised at how well it does, and how friendly it is.

Chats Sidebar for quick switching

One request that I started getting recently was right after our RocketHub launch. A handful of users from that community were asking for a sidebar where they could view a list of all their chats, just like in ChatGPT.

The minimal approach I took with Magai made Chats more akin to traditional document editors. And since most of us are used to a flow of clicking File > Open in tools like Google Docs or Microsoft Word, this seemed the natural setup for Magai. I wanted the UI to be both familiar and new, but with a focused, clutter-free page.

After some careful thought, I realized that I could maintain the beautiful minimalism, familiar workflow of opening files, and also add in a sidebar element that would allow you to view and quickly switch between chats.

This new Chat List allows you to view, search, sort/filter, and quickly switch between chats with ease. And when you don’t need it, it’s safely tucked away under the utility bar.