CRAZY Week In AI: New Models (1m Context Window)

What a crazy week.

In the last 7 days, we got new models from Mistral, Google, OpenAI, Cohere, an announcement from Meta, and the new music generator Udio. There’s so much going on it’s hard to keep track!

Keep reading to learn more about the new models and when you should be able to use them in MindStudio.

🗞️ Industry news

1. Gemini 1.5 Pro Now Available in 180+ Countries
   

   

   
   Gemini 1.5 Pro is finally available through API in 180+ countries. Previously, the only way to use 1.5 Pro was through the AI Studio, currently only available in the US.
   
   The new Gemini model shows awesome potential, and debuts with the first-of-its-kind 1 million context window. While Anthropic claims the Claude 3 family should support 1m in context, Gemini 1.5 Pro is now the only publicly available model that actually does.
   
   This means you could analyze a full 1-hour long video and get a summary with nearly perfect accuracy in retrieval… in one prompt, without using RAG.
   
   That being said, please note Gemini 1.5 Pro is not cheap during preview. Summarizing a 1-hour long video could cost around $7-8.
   

2. OpenAI releases update to GPT-4 Turbo & Function calling

   GPT-4 Turbo now has a dedicated API for function calling and has reclaimed the crown as state-of-the-art LLM in the arena.
   

   

   
   OpenAI claims there’s a substantial improvement, but doesn’t go into detail. From a few initial tests, it does appear to return better answers, but it’s too early to confirm whether it’s better than the previous GPT-4 Turbo.
   
   The model is now available in ChatGPT Plus. ChatGPT was previously using GPT-4, with a very limited context, in the ChatGPT Plus subscription.
   

3. Mistral releases new open source model: mixtral 8x22b
   Mistral has a weird tendency of posting random magnet links on X (previously Twitter) whenever there’s a new model ready.
   

   

   A user downloaded the full 281gb model on its Mac

   
   Shortly after release, Perplexity started offering mixtral 8×22b in their playground and hosting providers like DeepInfra now offer it through the API. It’s a very large model, but still not comparable to GPT-4 in parameters size.
   
   We’ll know how this performs in the arena soon. From initial user feedback, it looks like this is not a tuned model, so you’ll definitely need to provide examples of the outputs you’re looking for when using it.

4. Cohere announces ReRank 3 - new model for Enterprise Search
   Cohere entered the arena and matched the old GPT-4 in performance with its Command R+ model. Now, they’re entering the Enterprise Search area with ReRank 3.
   

   

   Cohere displays token optimization with its ReRank 3 compared to industry leading alternatives.

   
   The model is particularly good at multilingual outputs and token optimization, making it a potentially viable choice for the enterprise.

🔥 Product Updates

MindStudio is keeping an eye on the new models and will release access to most once a stable pipeline is available. We’re looking into the updated GPT-4 Turbo, Mixtral 8×22b, and Gemini 1.5 Pro right now.

Expect to see many of the new models in the platform soon. In the meantime, keep an eye on the LLM arena to estimate performance.

This week, MindStudio released:

• Implementation partners: we compiled a list of certified freelancers and agencies to help you get more done in less time. If you don’t have the time to learn how to use MindStudio or prefer to delegate the build, we suggest starting from here;

• Hackathon winners: we announced the MindStudio Hackathon winners in our Discord channel. Congratulations to the winners!

• Budgeting tools: more budgeting tools, such as limits by app and by users, are now available to all users;

• New usage dashboard: this will provide more details to users looking to understand where their funds are being spent.

Coming soon:

• Desktop app: MindStudio is working hard to deploy the application wrapped as a desktop app. This way, you’ll be able to offer it to your employees when they boot up the work laptop for the first time and use it on the side while working. We’re aiming for this month to release;

💡 Tip of The Week

You can use the “resolved” message in the debugger to test multiple models in the profiler.

In the debugger, each “programmatic” and “background” message includes two values:

• Sending Message: this is your prompt as is, with the variables in curly brackets and the instructions;

• Resolved Message: this is your actual prompt, aka what’s being sent to the AI. This matches the sending message but executes a function to retrieve the values in the variables, so it will expose what’s inside them.

If you copy the resolved message and paste it in the profiler, you can effectively check how other models would respond to the final prompt you send in the workflow, letting you test the active automations rather than a chat message.

🤝 Community Events

We’re working on a new set of workshops to offer in the next 2 weeks. We’ve heard your feedback loud and clear, and we’ll focus on:

• Live builds: we know you love to see MindStudio in action, and we’ll host two new live builds where we will walk through a real application and build it with you;

• Data sources: RAG is essential to building AIs infused with knowledge. Our first workshop was a success, and we’ll run another one to focus on more detailed examples;

• Q&A: how do you go from beginner to maestro? With tips and tricks coming from months of experience and well over 100 AIs built on the platform.

Here’s where you can register to the upcoming workshops:

How to Deploy AI Applications Everywhere using MindStudio and Zapier
04/18 10:00 AM PST
https://zoom.us/meeting/register/tJAsd-yoqjopHdOQnI5vOP2x9TBoS7oK-O3w

How to Use Your Own Data in MindStudio 
04/23 12:30 PST
https://zoom.us/meeting/register/tJwtc-qrrDsuGNMyjLOf8uBUKgR4sNyCiFaE

Learn How to Build a Content Generator (live build)
04/30 12:00 PM PST
https://zoom.us/meeting/register/tJArdO-prjwsHddSmu84Duo__uWpzbV9dk_j

Learn How to Build a Meeting Summarizer
05/09 10:00 AM PST
https://zoom.us/meeting/register/tJcpcuCsqzsjHNR13rnfYb85JOEiHLIubXGw

We will publish the other dates and topics on our blog.

🌯 That’s a wrap!

AI is moving faster than usual, and we’re happy to be part of this groundbreaking revolution. Stay tuned to learn more about what’s next and get tips & tricks for your MindStudio build.

You saw it here first,

Giorgio Barilla
MindStudio Developer & Project Manager @ MindStudio