What a month for AI...

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

Here at MindStudio, we’re working on making our API product amazing. Together with the new UI, you’re going to see many new updates next week 🙂 

In other news, the AI space has been pretty dormant for the past weeks until last Monday, when all companies decided to push groundbreakings news at once.

xAI now has a public API for Grok. Suno added image and video prompting with Suno Scenes. Pika launched a plethora of new video effects to their generative video product. ChatGPT introduced Canvas. Anthropic revealed Claude 3.5 Sonnet New, Claude 3.5 Haiku, and Claude Data Analysis to execute code in the chatbot.

Continue reading to learn more!

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

🗞️ Industry news

OpenAI focuses on its cash cow: ChatGPT

OpenAI makes a significant majority of its revenue from ChatGPT. Some reports estimate up to 80-85% of OpenAI’s revenue comes from ChatGPT Plus, Team, and Enterprise, with most of the Fortune 500 US companies reporting using it for AI tasks.

Canvas is another step in the consumer direction, offering the winning solution by Claude Artifact, improved for long-term use.

ChatGPT Canvas lets you:

• Work with AI side-by-side, abandoning the original chatbot interface. This is much better when working with large documents, for example when writing an article or editing a codebase, and comes with beautiful animations;

• Canvas is currently using GPT-4o, not the latest o1 family. While GPT-4o has a lower max output tokens count, that seems to work better in the special model used in Canvas. This is not the standard GPT-4o but a specific version tuned for long format documents;

• Personally, I’m loving Canvas when working on articles or things that require polishing the AI response. Instead of rewriting the whole thing, ChatGPT Canvas can effectively edit documents and add small components without messing up with the overall context;

• The sidebar interface simply works better on large screens, versus trying to convince the chatbot to output large responses.

With ChatGPT Canvas and the recent Voice Mode (now available to all users in the EU as well), OpenAI seems more focused on consumer tech than ever.

That makes sense. My nephew is 15 and confirmed teens now say “ask Chat” like we used to say “Google it” - and when your product is so common it becomes a meme or a verb you know you’re doing something right.

But, the voice commands don’t seem to appeal as much as chat. Was it just first mover advantage or is there another product ChatGPT could ship to reach higher profit margins?

Anthropic releases Claude 3.5 Sonnet New & Wants to Control your Computer

This week, Anthropic released Claude 3.5 Sonnet New (already available in MindStudio) and hinted at Claude 3.5 Haiku, coming this month. 3.5 New scores better than any other model other than the o1 family.

Claude 3.5 Haiku is particularly interesting, as it’s supposed to outperform the old Claude 3.5 Sonnet in coding. This also marks the end of an era for Opus, the most expensive model. While Anthropic promised a release, it never came, and the big model is now practically meritless compared to the new 3.5 Sonnet and the upcoming Haiku.

Together with the new models, Anthropic shocked the AI world with “Computer Use”. The idea is simple: why build “agents” if you can teach the AI to use the computer and let it do what it wants on the screen?

Computer Use basically takes a screenshot of whatever you’re looking at, identifies the possible actions, maps them to the task at hand, and moves step by step to reach the goal. It’s trained on the most common tools used in businesses.

And yes, of course I tried it. I wouldn’t be a good nerd if I didn’t. You can try it here on a Replit template (no coding knowledge required).

First impressions:

• This is very, very slow. It takes forever to execute the most basic tasks. This is somewhat expected, but still odd the first time you try it;

• It’s very impressive. Seeing an AI control whatever you see in front of you is just a mesmerizing experience. There were similar tools in the past, but none came close to the great performance Anthropic managed to get with Computer Use;

• The new mode is very expensive to run. I used it for a few minutes, and it used over 100,000 input tokens. It’s not clear how this will be priced either. Treat it as an alpha/beta product, as stated, and don’t use it in production.

Anthropic has a healthier portion of revenue coming from the API platform, which might explain why they’re actually shipping more ambitious features compared to OpenAI.

As a Thursday gift, Anthropic also announced Claude AI will now be able to execute code, mimicking the existing “Code Interpreter” functionality embedded in ChatGPT. You can now run Javascript in a sandbox while using Claude.

Generative AI in Videos is… good?

Until very recently, AI videos were bad. Yes, they were impressive. Yes, they were fun. But they were bad - nothing you’d use on a daily basis, and nothing you’d probably pay for.

That might be changing with Runway Act One.

Runway Act One can create compelling animations using video and voice performances as inputs. It represents a significant step forward in using generative models for expressive live action and animated content.

Act-One can be applied to a wide variety of reference images. The model preserves realistic facial expressions and accurately translates performances into characters with proportions different from the original source video.

But this is corporate lingo, isn’t it? Take a look at the dozens of examples posted by Runway here. They’re good, and mark a milestone: hybrid AI video generation.

Just like AI voices aren’t 100% there, but voice cloning is very close and used daily in software like Descript (my videos sometimes feature AI Giorgio, and you probably can’t say where). Act One can bring the same concept of voice cloning to video.

Instead of using Generative AI for the whole thing, you’d use it to complement your human intuition while waiting for a model that can genuinely, realistically, create assets you can use. That seems like a great idea to test!

🔥 Product Updates

Users can now add sticky notes in the workflow builder. Notes can have a header, use a plethora of colors, fill any size, and are permanent until removed.

Use notes to:

• Understand complex workflows in a few minutes

• Group multi-modal blocks in one color and add an header to identify what that group of blocks does

• Make it easier to collaborate with multiple team members

• Effectively add "comments" to your canvas like you'd do in code

Notes are a further step to full collaboration in MindStudio apps. Together with the option to "take over" editing for an app, this makes it even easier for large teams to work together on AI processes.

In addition to Notes, we shipped a few updates to the API product and quality of life improvements:

• API now outputs a “result” at the end, without having to fish for the right ChatMessage output in an array. This makes it easier to call a MindStudio flow and get the output;

• The “Run Workflow” block is a new block to execute a MindStudio workflow, in the same app or any other in your account. Once you build your flow once, you can now reuse it across all apps;

• We keep upgrading all models in the platform, and released Claude 3.5 Sonnet New hours after release. We will release Claude 3.5 Haiku as well, and it will become the new default model in the platform.

💡 Tip of The Week

OpenAI and Anthropic are showing how much products like MindStudio are needed in the market for builders.

The biggest model provider in the world, OpenAI, makes more money from the playground they built to test the API than the API itself.

There’s never been a better time to build AI solutions and interfaces for your clients, and MindStudio can help you do it at lightning speed. The market is now:

• Old enough. We’re 2y in, and people are aware of what AI is, what it can do, and the potential to change how they work;

• Less scared. We got that AI is very powerful, sometimes scary, but most agree it’s not going to end the world. There are less articles talking about how it’s going to end us all;

• Money oriented. Once we got past the VC stage, now companies are trying to find ways to make money. If one of the largest players chose interfaces, that means our focus of giving you the tools to build products around your market is the right choice to increase your revenue.

Look, building another ChatGPT doesn’t make much sense. We’re going to continue building a better chat endpoint, but our end goal is enabling you to build custom solutions for your clients. Things people are happy to pay for.

We want to empower you to partner up with MindStudio and use the tools to build niche-dominating workflows and apps.

If you have any suggestions on how we can do this better, please reply to this email! I read all replies 🙂 

🤝 Community Events

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 🔥

If you’re interested in any topic in particular, feel free to reply and I’ll do my best to include it in the next releases. We’re going to update all of these soon.

🌯 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