BB Digest: A peaceful start to 2025

Digest #521 → Subscribe | Upgrade to Pro

Hey folks, Happy New Year (unacceptable to say on the 7th?! 😬)

The holiday season was surprisingly civil—every major lab respected our sanity with measured updates (thank you!). But don't get too comfortable. 2025 is going to be massive, particularly for AI agents. So I've packed today's digest with a little extra reading material.

TLDR; inside today’s newsletter

• OpenAI’s next model, Google thinks and scrappy wins from China

• ICYMI: new tutorials for product analysis and content widgets

• snack n skill - build your first full-stack web app with AI

• 7 new AI tools and updates

• 9 posts exploring agents, learnings and predictions

• to-do’s

Alright, let’s get to it!

 🔎 Trends & news

• A recap of the last days of 2024:

  • OpenAI teased a new family of models o3 and o3-mini. These models cross the expert threshold in many cases and “solve” the ARC-AGI benchmark (read more). Also, ChatGPT’s Desktop app can now work with Notion and more code editors.

  • Gemini Flash 2.0 Thinking - Another experimental model from Google that “thinks” like OpenAI’s o1 models. Though Google has decided to show the model’s thoughts, unlike OpenAI.

  • More people got access to Google’s Veo 2. Content on my Twitter timeline was 10:1 in favour of Veo 2, despite OpenAI’s offer of unlimited Sora usage for Plus users during holidays.

  • Deepseek V3 - A new model from China shocked everyone by training a near-SOTA model with a fraction of the resources. The model is open-source with a rich paper describing their work.

• Sam Altman has a new blog reflecting on 9+ years of OpenAI. Sam claims that OpenAI now knows how to build AGI by their definition and wants to aim higher for ASI - Artificial Super Intelligence.

• In Jan we can expect stable releases of Gemini 2.0 Flash models, o3-mini and likely Grok 3 too. Anthropic has been quiet, but they often ship without building hype. It’s going to be an exciting start to 2025.

• A call for early-stage B2B SaaS founders! Secure an up to $250K investment from Forum Ventures in 2025, plus expert hands-on support, fundraising guidance, and investor connections. If you have an idea, an MVP or product in-market, pitch them today!*
  *sponsored

👀 ICYMI: New stuff we’ve launched

Tutorials

• Automatically analyze product reviews and feedback

• Add a ‘talk to this article’ widget to your blog

Blog

• How to bring your product to market with AI

🍽️ Snack ‘n’ skill

Learn how to build a full-stack web app without coding

Ever wanted to build your own web app but got stuck at "Hello World"? AI-powered tools have finally made it possible for non-coders to create functional web apps. This tutorial shows you how to build a Kanban-style todo app using Lovable and Supabase.

Step 1. Create your app interface

Start by heading to Lovable and describing your app in plain English. A simple prompt like "A fullstack to-do list app with a drag-and-drop Kanban board" is enough to get started. Lovable will generate both the interface and code automatically.

Step 2. Make it fully functional

• Set up a free Supabase account and create a new project for your database

• Connect your Supabase project to Lovable using the integration button

• Let Lovable generate the database tables needed to store your todo items

• Test the app and use Lovable's "Try to fix" button to resolve any errors

• Refine the UI by sharing design inspiration images with Lovable

• Deploy your finished app by clicking "Publish" and then "Deploy"

Tip: Start simple and test the core functionality before spending time on design improvements. It's easier to iterate on the look once everything works.

This walkthrough is a condensed version of this tutorial. Check out the full tutorial for the complete process with screenshots.

⚙️ Top new tools

• Fuel iX: One platform to manage and connect your infrastructure with 20+ premium GenAI models while maintaining full tech stack control.*

• Margins - Track your reading and search by vibes. (read more)

• Style models by Flair AI - Train models on any particular aesthetic.

• Shortest - Write tests in plain English and let AI handle the execution.

• Gemini Search - Perplexity clone built using Gemini 2.0 and Google search.

• GeminiCoder - Use Gemini to build an entire app.

• Claude can now analyze large Excel files up to 30MB.

More tools here →
*sponsored

 📜 Interesting posts

• A guide for building effective agents from Anthropic.

• OpenAI has new plans for its profit structure.

• ‘The stakes are high’ - Sundar Pichai to Google employees.

• Beginner's guide to building a website using AI.

• Travis Scott X Nike commercial used 5000+ images generated with Midjourney for inspiration.

• Things we learned about LLMs in 2024.

• 3 predictions for the future of AI agents in 2025.

• AI clones are good enough now.

• Learn to build with multimodal AI (for technical users).

📌 To-dos

1. Let us know if you have any tutorials or courses you want us to create—we’re always open to ideas.

That wraps things up for today! See you again on Thursday. 👋

Ben