GPT 5.5 passed the ultimate intelligence test: hacking proprietary hardware
The creator reverse-engineers a Divoom Mini 2 retro Bluetooth speaker/screen by using a packet sniffer to capture Bluetooth traffic from the official app, then feeds those logs to GPT-5.5 (via Codex) to decode the protocol and display custom content on the screen.
Summary
In this short video, the creator showcases a Divoom Mini 2, a retro PC-style Bluetooth speaker with a small pixel display. Their goal is simple but technically interesting: bypass the official app and display custom content on the screen programmatically.
To accomplish this, they connected the device to a packet sniffer while using the official phone app to send images to the device. The sniffer captured the raw Bluetooth packets being transmitted, giving them a log of the proprietary communication protocol.
Those Bluetooth logs were then fed into GPT-5.5 (accessed through OpenAI's Codex), which analyzed the packet data and helped decode how the device communicates. The creator built a custom notify hook into their Codex configuration, so that when an AI task completes, it triggers a notification displayed on the Divoom screen.
The demo shows this working in real time: the creator types a simple prompt in the terminal, and when the response is ready, the Divoom screen displays a custom message ('Your move' — with a playful note that the 'e' was missing) and plays a beep sound. The creator expresses genuine excitement about the result, calling it life-changing.
Key Insights
- The creator used a packet sniffer to intercept Bluetooth traffic between the official Divoom app and the device, treating the proprietary protocol as a black box to be decoded rather than documented.
- The creator fed raw Bluetooth packet logs directly into GPT-5.5, demonstrating that LLMs can be used as protocol analysis tools for reverse engineering undocumented hardware communication.
- Rather than a standalone script, the creator integrated the Divoom display control into a Codex config as a notify hook, so the hardware screen becomes part of an AI coding workflow.
- The live demo confirms the system works end-to-end: a terminal prompt triggers Codex, which on completion pushes a custom message to the physical Divoom screen and plays a sound.
- The creator frames this outcome as genuinely transformative ('this is changing my life'), suggesting that using AI to decode proprietary hardware protocols dramatically lowers the barrier to hardware customization.
Topics
Transcript
[0:00] My real e-thou is this thing. This is a Divoom mini 2 retro PC style Bluetooth speaker and tiny [music] screen. And my only goal is to be able to display funny stuff on this screen. I then hooked it up to a packet sniffer so that when I was using the app here on my phone and it sent an image to this computer, it would log and sniff the packets and tell me what Bluetooth was sending to this little guy. And I threw those logs at 5.5 and let me show you what happened. [0:32] So, I have my terminal up and I'm going to say something really simple. What can you help me with? And…
Full transcript available for MurmurCast members
Sign Up to AccessMore from How I AI
Muse dodged every trap in this billing ticket
A billing ticket resolution task was used to benchmark agentic AI performance, and Muse excelled by successfully navigating multiple challenges including account linking errors, duplicate refund issues, and prompt injection attempts. The agent demonstrated superior performance by investigating thoroughly, fixing the account link, issuing the correct refund, updating the plan, and marking the ticket as resolved.
Meta's Muse is the best AI agent I've tested
The speaker provides their initial impressions of Meta's Muse AI agent, declaring it the best agentic AI they've tested. Muse excelled across four different knowledge work tasks including billing management, customer responses, inbox triage, and research, outperforming alternatives.
Meta's Muse: best AI agent, worst designer
The speaker tests Meta's AI model on video editing and finds it excels at technical tasks like cutting and timing but significantly underperforms on creative work like design and overlays. They conclude that while the model is strong for coding and agent tasks, it falls short compared to Claude Opus 3.5 for creative applications.
No two Muse agents are the same
The speaker discusses how Muse agent setups are highly personalized and variable, emphasizing the importance of understanding underlying system architecture rather than obsessing over individual details. They describe adopting a more flexible, liberating approach to design that accepts good outputs without requiring perfectionism.
Is Meta’s Muse Wonder model any good?
Clare Vo reviews Meta's Muse Wonder AI model and finds it excels at agentic work and coding tasks but significantly underperforms on creative design work compared to competitors like Claude Opus 3.5. The model shows promise as a personal AI agent with strong domain expertise and memory capabilities, but lacks the design sensibility needed for front-end development.