BB Digest: Is OpenAI open again?

Hey folks, no new post from me today, but ICYMI, I’m back at creating videos to make building with AI easier. So far, I’ve covered:

• GitHub & Git: for non-technical founders.

• Creating an extension to supercharge Bolt with "Cursor Rules"

• A head to head battle between 4 AI coding tools.

check em out and let me know what you want next.

TLDR; inside today’s newsletter

• OpenAI’s attempt to be open again and Amazon’s secret browsing agent

• new video generation models in the market

• words from today’s sponsor: Speechmatics

• updates from Manus AI, Gemini Coder and more

• a new owner for Twitter

 🔎 What’s Trending

• OpenAI plans to release an open-weights model soon. They are looking for feedback from the developer community to make it worthwhile. It’ll likely be a large model that can still run on a single GPU (instead of a model that you can run on your mobile). The goal appears to be to keep companies and governments that want local AI on their side.

• Amazon gave us a sneak peak into their version of a browser using agent. They call it Nova Act. It’s currently available as a toolkit for just devs but it seems to work better than OpenAI’s operator or Anthropic’s Computer Use.

• RunwayML has a new video generation model: Gen-4. They are focusing on making video generation controllable so that it can be used in actual workflows. That seems to be a common theme because another AI video generator, Higgsfield, is excelling at replicating camera movements in its outputs.

• Frustrated when AI doesn't understand you?

  Speechmatics delivers 25% more accurate real-time speech recognition with 90%+ accuracy across 55+ languages. While others rush to make AI talk, we've solved what matters: making it truly listen. Your AI should keep up, not catch up. Try it free today.*

• Eleven Labs have trained a model for dogs…no, seriously. Simply type a message, choose your breed, and their models will convert it into fluent barking. With independent benchmarks stating 95% of dogs couldn’t tell the difference!
  *sponsored

⚙️ Tools and News

• Your AI should keep up, not catch up Speechmatics: 25% more accurate real-time speech recognition that gets you the first time. Try it free.*

• Native image generation in ChatGPT is now available to all free users as well (3 images per day). Here are 16 other ways to use ChatGPT image gen other than making Studio Ghibli style images.

• Gemini 2.5 Pro is now also available to free users of the Gemini App. It integrates with Canvas and can create games that run in the browser. I personally find reading its thoughts the most delightful.

• Google is wondering “To MCP or not to MCP” as it gets adopted by the AI community. There’s even one for WhatsApp now.

• Mirai is helping developers build applications that use on-device AI without the ML shenanigans. Easily enable smart input, auto-reply, semantic search, and more features with AI that runs on your iPhone—no cloud, no GPUs. I am an investor.

• Developers love Claude Code - A command line tool for writing code. The PM behind NotebookLM built a version that uses Gemini instead. Gemini Code uses Gemini 2.5 Pro in the background, and in my little testing, this model is really better than Claude for coding.

• Manus AI, the viral browser using agent, now has pricing plans: $39/mo and $199/mo. I predicted that in my unboxing of Manus.

• PocketFlow lets you use AI to build AI workflows. Their dev framework is open source, but you can use their platform if you don’t wanna code.

• Scene is a modular AI web builder that combines the flexibility of drag and drop with the ability to create images, copy and wireframes with AI. It’s not a coding tool, but a modern take on website builders like Squarespace.

• Elon Musk's AI company, xAI, is now the owner of X. It values xAI at $80B and X at $33B. I’ll say it’s not entirely surprising. Twitter was already xAI’s private data bank.

• AI therapy seems like it’s working. A 200-person clinical trial at Dartmouth used fine-tuned Llama and saw depression symptoms drop by 51%, anxiety by 31%, and eating disorders by 19%. It’s still early days, but good signs.

• Anthropic released an update to its work on the economic use of Claude, comparing the changes in key categories. And in typical Anthropic fashion, they also released another paper, tracing Claude’s thoughts to understand its biology.

Our free newsfeed for more →
*sponsored

That’s it for today. Feel free to hit reply and share your thoughts. 👋