Model Mixer & MindStudio v2 Debut; OpenAI Voice Mode Postponed

You're receiving this email because you registered to one of our workshops. You can unsubscribe at the bottom of each email at any time.

This week, MindStudio released two new features together with the Claude 3.5 Sonnet model addition: Model Mixer and extended team collaboration features. I also upgraded our internal trainer to version 2, with several quality of life improvements and new workflows.

The industry also moved forward, with Anthropic being the biggest winner this time of the year. Claude now features “Projects” (basically, GPTs), Meta is testing AI Studio with a few select influencers, and OpenAI disappointed everyone postponing Voice Mode to fall 2024 while releasing an incredibly underwhelming MacOS app for ChatGPT to all users.

As a reminder, we’re now welcoming partners that want to build AIs for their clients. Sign up for extra support, training resources, and more here.

🗞️ Industry news

1. Anthropic launches “Projects”, its play for teams
   
   

   

   Copyright Anthropic

   
   
   Anthropic has launched Projects for Claude.ai Pro and Team users, introducing a new way to organize and collaborate within the platform. The feature allows users to curate knowledge and chat activity in one place, with a 200K context window and powered by Claude 3.5 Sonnet. Basically, GPTs for Claude.

   
   A few highlights shared by the team:
   

   • Grounding Claude in internal knowledge: Users can upload documents, code, and other relevant information to enhance Claude's effectiveness for specific tasks;

   • Artifacts: A new feature in preview that allows users to generate and interact with content like code snippets, documents, graphics, and website designs in a dedicated window alongside conversations;

   • Sharing capabilities: Team users can share snapshots of their best conversations in a shared project activity feed, fostering inspiration and skill improvement.

     
     

   I’ve been personally enjoying the Artifact feature, and there are a plethora of startups trying to integrate it. MindStudio is considering new output modes which might resemble Artifacts. Although we can’t make promises right now, it’s something we’re looking into!
   
   Claude.ai still can’t search the internet. You can easily build an AI that can do that and much more, without relying on only one model, in MindStudio.
   
   

2. OpenAI delays GPT-4o voice and releases underwhelming MacOS app
   
   

   

   
   
   It genuinely feels like OpenAI is struggling to innovate at this point. With Anthropic, Meta, and Amazon moving at full speed, OpenAI is now googling their way through new features.
   
   Just like Google, OpenAI is now a company that announces a new feature, makes it look like they actually have it, and then asks you to patiently wait 6 to 12 months while they actually make it ready for the public.
   
   After announcing a version of ChatGPT that can screen record your workflows and talk to you in real time with GPT-4o voice in May 2024, they now released a ChatGPT app that can barely take a screenshot for you and delayed voice to fall 2024, citing security concerns.
   
   Mira Murati (CTO) and Sam Altman (CEO) are also hosting live interviews making it clear GPT-5 is probably going to take much longer to release than previously expected, even though the company confirmed it’s training the next frontier model.
   
   That being said, all companies have ups and downs, and it’s why we always say multi model is better. Relying on one provider means missing out on the innovation other players are bringing to the table, like the extraordinary coding skills of Claude 3.5 Sonnet, now scoring over 1300 ELO in the Arena, higher than GPT-4o.
   
   OpenAI is still doing some work to improve their models like partnering with TIME, training AI to troubleshoot AI errors, and a new cybersecurity grant. I’d love to see them act more like Anthropic, though, with a history of releasing features just after the announcement. Both Claude 3.5 Sonnet and Claude Projects were live minutes after their blog went live.
   
   

3. Meta starts testing AI Studio with select Instagram creators
   
   

   

   
   Meta has begun testing user-created AI chatbots on Instagram in the U.S. These chatbots, developed through Meta AI studio, will be clearly labeled as AI and primarily appear in messaging.
   
   The rollout starts with about 50 creators and a small percentage of users, with plans for full launch by August. These will be powered by Llama 3, a very capable open source model.
   
   In a recent interview, Mark Zuckerberg (CEO) said he doesn’t believe there will be a sort-of mastermind AI, but that people will interact with all sorts of different models for different tasks, just like they now use multiple apps for different use cases. Listen to the interview here.
   
   While the concept shows promise, it's worth noting that Meta is not the first to explore this space - a16z-backed Character.AI is simultaneously launching AI avatar voice calls, and people have been using platforms like MindStudio to create AI personas for months now.

In other news, there are some rumors of a new AI chip able to output 500,000 tokens per second running Llama 3 70b (pure speculation at this point), OpenAI acquired a collaborative coding IDE, and Apple is the first company to be investigated under the EU’s DMA regulation.

I guess that might explain why the company won’t release the hyped Apple Intelligence to EU consumers in 2024. Personal opinions aside, the DMA regulation is very abstract, and it will make it harder to release AI to big platforms. The EU currently recognizes the following as “gatekeepers” under the DMA regulation:

• Alphabet: Google Maps, Google Play, Google Shopping, YouTube, Google Search, Chrome, Google Android, Google

• Amazon: Amazon Marketplace, Amazon

• Apple: App Store, iOs, Safari

• ByteDance: TikTok

• Meta: Meta, Facebook, Instagram, WhatsApp, Messenger, Meta Marketplace

• Microsoft: LinkedIn, Windows PC OS

🔥 Product Updates

MindStudio released enhanced team collaboration features this week. Team members can now edit each other AIs and “kick” other members out of the editor to take over.

We also released the Model Mixer, which lets you instantly compare responses from up to four models simultaneously within the chat. You can also mix and match unlimited models in one thread. This is a UI-facing iteration of the Profiler, one of our best features to figure out which model you should choose for your next AI build.

Together with product updates, the team and I worked on upgrading our internal education app: MindStudio Trainer.

MindStudio Trainer v2 is here, and comes with awesome updates:

• [New] Learning Path Builder: generate step-by-step guides for anything you want to learn. The AI will fetch the right links from our knowledge base, YouTube tutorials, and events.

• [New] Data Source Builder: build your own data sources with MindStudio Trainer v2. You can now scrape your website or YouTube channel with one click and get the CSV output to upload in MindStudio.

• Lots of quality of life improvements: all workflows are now cheaper, faster, and better. You'll get great results, especially from the custom function and right model workflows, and notice you'll be more productive while building with MindStudio.

You can use it by creating a new app - it's the first template in the list. Learn more here - the landing page includes FAQs, highlights, and a video overview.

Remember, the MindStudio trainer is an app like any other, which means you can reverse-engineer all workflows and upgrade them with your data sources and prompts.

💡 Tip of The Week

Courtesy of power user @argyrisk in Discord, MindStudio can now be used to generate graphs for any data source.

Quickchart.io is the secret sauce behind the graph generation capability, making it possible to generate charts with a JS library (ChartJS) by encoding the values in the URL and generating an image you can then display in Markdown.

Quickchart is free to use and works great in MindStudio, granted you’ll need to work on your prompt. You need to tell the AI what you want to achieve, provide the data table, and prompt it to generate the QuickChart URL. Then, you need to display the image with markdown.

Here are the basics:

• Don’t rely on the AI layer to understand how to generate a quickchart.io chart. Provide snippets of the documentation and add example outputs for all chart types you expect it to generate;

• Ensure you add the axis labels to all examples, otherwise the chart will be hard to read;

• Use a smart model like Claude 3.5 Sonnet or GPT-4o, smaller models might struggle to follow the right syntax or understand you need the final image URL, not a chart using ChartJS;

• Avoid using very large data tables. Remember you’re going to upload the data with the file upload input type, meaning MindStudio will add your whole data table to the prompt. If your data table is very big, then the prompt cost will be very high as well.

If you want a boilerplate, here’s a demo app I built. It works for simple graphs, but might return errors for sankeys and complex chart types. You can clone it in your workspace or simply copy the prompt and use it in your workflows.

🤝 Community Events

If you want to hangout with our team, we usually host a Discord event every Friday @ 3PM Eastern. Join our Discord channel to keep up to date with the hangouts - our entire team is active there.

You can register for upcoming events on our brand new events page here.

Our new webinar series is up on there as well, with the following on-demand webinars:

• MindStudio for Partners: Level Up Your Agency With AI Workflows

• Customize AI Applications With Your Data

Plus, we have new weekly and bi-weekly events:

• How To Build Your First AI Assistant (the best webinar for beginners!)

• Which Models to Choose for Your AI

• First Steps to Bring AI Into Your Workflow

Thank you for being an invaluable member of our community, it’s always great to see many of you join multiple workshops 🔥

🌯 That’s a wrap!

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