Скачать или смотреть This $70 AI Chip Can't Run LLMs... So I Made It Do This Instead

In this video, I detail my journey of experimenting with a $70 AI chip, thinking I could run large language models (LLMs) on it. While that didn't pan out as expected, I discovered a far more efficient use for the chip in audio processing. Building on my previous project involving a Raspberry Pi with microphones and a speaker to create a smart voice assistant, I faced challenges including audio latency and interference from the speaker output. By integrating the Hailo AI chip and its optimized transcription models, I achieved localized audio transcription, enhancing both performance and privacy. I also explore other hardware options like the Nvidia Jetson Nano and Orange Pi, discussing their pros and cons. Lastly, I introduce a ReSpeaker hat for solving some wiring issues, aiming to create a streamlined, local AI assistant. Watch to see the evolution of this exciting project and share your thoughts and feedback!

00:00 Introduction: The $70 AI Chip Experiment
00:09 Previous Project Recap: Raspberry Pi Voice Assistant
00:29 Challenges with Audio Processing
01:56 Exploring the Hailo AI Chip
02:34 Local Processing with Hailo AI and Raspberry Pi
03:41 Jetson Nano: An Alternative Approach
05:21 Orange Pi: A Potential Contender
05:43 Future Plans and Conclusion