Hey, happy Tuesday. It's a great day to read the Create With newsletter! ✨

And welcome to the 371 new subscribers 👋

We're here to summarise the latest in AI news and give you practical tips and ideas to get creating!
​
Flip open your notebooks and let's get those ideas flowing 📝

🔎 Preview: In today's issue

• 🦙 Meta release Llama 4 - a Deepseek competitor with bells on
• ❓ What 'multi-modal' models mean for makers (try saying that 5x fast)
• 🎙️ Decoded: How to build a second brain with n8n and Pinecone
• 🤑 More hackathons to join and $$$ in prizes

😣 Explain like I'm 5 (Jargon Buster)
​

With Meta announcing its Llama 4 family with multimodal capabilities and industry-leading context length we thought we'd break down what the term means for builders.

Simply, it means “many ways of communicating”. It’s like how humans use voice, gestures, drawings, and writing to express themselves—AI can now do that too.

Some even call it “omni-modal,” which means all types of inputs and outputs are in play. For example the "o" in GPT4o stands for "omni" (fun fact).

However, your experience in ChatGPT is different if you are trying to build with their API. Think about a voice based app for a second:

✅ What you can do:

Send audio to /v1/audio/transcriptions​
→ It uses OpenAI’s Whisper model to convert speech → text.

Use the transcribed text in /v1/chat/completions​
→ Just like regular prompts.
​
(the above urls are API endpoints)

🚫 What you can't do (yet):

Send a raw .mp3, .wav, etc. directly to /v1/chat/completions and have a live voice chat through the API.

Get voice input and output in a single API call still requires separate steps so the more multimodal an LLM the more features it has for you to use it in your application.

More about Meta's release in the next section 👇

🔊 On your radar

Keep your finger on the pulse on what's hot in the space right now

⭐️ Meta is trying really hard to have a Deepseek moment with its Llama 4 open source "multimodal" release. According to lmarena.org it's the #1 open model, surpassing DeepSeek and Tied #1 in Hard Prompts, Coding, Math, Creative Writing.

👉 This matters because it's super powerful with a large context window e.g 20 hours of video or the whole Harry Potter series. And It's cheap; about as cheap as Google's Gemini series (if you use an open source api provider such as Groq).

But as it's open source you can also download and host the model yourself (more work and not for the faint hearted). Here's how to play around right now: It's available on WhatsApp, Web (Europe may have to wait) and Huggingface.

As you think about how to scale your product it's worth thinking about how to leverage open source models.

⭐️ Microsoft Copilot comes after consumers with an upgraded multi-modal user experience.

👉 Not quite a ChatGPT 4o moment unfortunately, but they're catching up.

For pure Tuesday vibes, Let's compare to OpenAI's 4o.

Notice how Copilot couldn't replicate the previous image. ChatGPT 4o did surprisingly well here.

⭐️ Bubble announce an AI Build Challenge with $3000 of cash prizes. Need we say more?

👉 Jumping on the highly successful campaigns by Lovable and Bolt.new; Bubble are going all out to capture some of the AI builder excitement. There will be 3 winners:

1. Most-liked app on X
2. Most-liked app on LinkedIn
3. Bubble's Choice award

Talking of hackathon's Lovable announced the winners from their hackathon. Our fav is the $10k winner Stardust Analytics app that helps Shopify store owners find the exact KPI blocking their growth.

Extra bytes

• In an epic crossover of music and tech, the World's Largest Hackathon will bring the Chainsmokers to DJ the opening party in SF! You can try to get tickets for this historic event here.
• 🚨 100 School's April lightning lessons kick off today! They have an amazing lineup of experts and all the sessions are free to attend! Make room on your schedule which includes big names like Zapier, Beehiiv, Bubble, Glide and more! ⚡️

🔬 Decoded: How to SUPERCHARGE your brain with n8n, Pinecone and Google Sheets

​

In this episode, we sit down with Ranjit Bhinge—founder of Atomic Fusion and no-code agency Blur Apps—to hear how he’s automating his LinkedIn content creation using AI agents, fine-tuned models, and personal voice journals.

He walks us through his full workflow using n8n, Google Sheets, Pinecone, and OpenAI—turning scattered thoughts into structured, authentic content at scale.

​Watch it here​

Get your Create With Conference ticket today!

Each ticket includes:

• a full day of keynotes, workshops, panels and insightful talks
• networking with 300 builders, founders and business teams
• demos of the latest AI features and platforms
• complementary tea, coffee + lunch
• access to the official afterparty
• and a couple more surprises 😉

Already got your invite for the 22nd of May in London? Keep an eye out for some special events coming to your inbox soon 🪩

📍 Final thoughts

We came across the interactive AI 2027 this weekend and it sure gave us a lot of thoughts...

📌 TL;DR

By 2027, we’ll see AI agents managing tasks, building software, even making strategic decisions. But amid this shift, two things remain crucial:

🧠 1. Human Expertise Still Matters​
AI can do a lot—but it can’t replace empathy, creativity, judgment, or trust.
If you’re a small business owner or freelancer, now’s the time to identify the parts of your work that only you can do.​
Then, look for AI tools that complement you, not compete with you. That’s where the real leverage lies.

📚 2. Continuous Learning Isn’t Optional​
The pace of change means skills have a shelf life.
To stay ahead, we need to be curious, adaptable, and open to experimentation.​
Start exploring AI tools in your current workflows and solving small challenges. Subscribe to updates. Take the time to learn—even a little each week goes a long way.

With good vibes 👋
​
The Create With Team ✨