Ep 251: Pluto, Pinball, Speedy Surgery, and DIY GPS

Hackaday Podcast · 5 Jan 2024 · 1h 06m · Episode 251

AI in practice: 2 real uses of AI described

What this episode covers

Hosts Elliot Williams and Dan Maloney review a diverse set of hardware hacks, including a DIY GPS receiver built from scratch by Leaning Tower, a Raspberry Pi-based system for localizing fireworks by Kim Hendrix, and a mechanical pinball logic analysis by Technology Connections. They also discuss a local AI tool by Justine Tunney for renaming images, a video feedback machine by David Blair, and a surgical efficiency model by Lewin Day. The episode concludes with a look at 2023 hardware trends, highlighting the rise of affordable PCBA services and faster 3D printing.

AI in practice

Where this conversation describes AI being used to build or deliver a product. Part of the AI in practice tracker.

Designlava (large language, visual model)4 seconds per image on a $8300 Mac with 60 GPU cores; 45 seconds to 1 minute on less powerful hardware

A local large language model analyzed images to generate descriptive file names, automating the renaming of image files based on their visual content.

“Justine Tunney … wrote up a large language model running locally to rename image files based on their appearance and to do it all automatically. … Using a bunch of Mozilla developed tools for local, large language models.”

Speaker 2 · Hear it at 37:10

Running the companyApple SpeechTranscriber

The podcast production team used an on-device speech-to-text tool to generate the transcript of the audio episode.

“method: "speech-to-text, on-device: Apple SpeechTranscriber (words) + FluidAudio offline diarization (speakers…”

transcript metadata ·

Mentioned

Companies
Cornell University, Mozilla, Open AI
Chips
Raspberry Pi, FPGA, CH32V003, Raspberry Pi Pico, MPU 6050
Tools
Raven Light, Scanimate
Also
Chaos Communication Congress, Pluto, David Blair, Leaning Tower, Technology Connections, Justine Tunney, Chat GPT, Lewin Day

Other shows on the same ground, closest first.

The summary, chapters and topics are generated from the episode’s audio by this directory, not written by Hackaday Podcast. The episode itself belongs to its publisher.