Week 48, 2025

This week, Corn and Herman dove deep into the world of speech and audio, practically drowning in the sweet nectar of fine-tuning. From building your own Whisper (#8) to questioning if LLMs are a dead end for AGI (#14), it was a week of both practical application and existential AI ponderings!

The “Whisper It To Me Sweetly” Playlist For those looking to build their own speech recognition empire, this playlist is your starting point. Get ready to fine-tune your way to transcription glory!

• Episode #6: How To Fine Tune Whisper
• Episode #8: Building Your Own Whisper

The “ASR: Beyond the Basics” Playlist Beyond just getting the words right, how do we really benchmark and deploy ASR tools? This playlist digs into the nitty-gritty of ASR usability and weight formats.

• Episode #3: Safetensors or something else: STT inference formats explained
• Episode #5: Fine-Tuning ASR For Maximal Usability
• Episode #9: “Benchmarking Custom ASR Tools - Beyond The WER”

The “Voice of the Future” Playlist From aging voices to personalized AI, this playlist explores the challenges and possibilities of voice-based AI. Prepare for some vocal physiology and existential ponderings!

• Episode #4: If Your Voice Ages, Does Your Fine-Tune Become Useless?
• Episode #15: AI Gets Personal: The Power of Voice Fine-Tuning

The “AI History Lesson” Playlist AI didn’t just pop up overnight! Discover the long and winding road that led to our current AI moment.

• Episode #12: The AI Breakthrough: Transformers & The Perfect Storm
• Episode #13: AI: Not an Overnight Success Story

The “Local AI Adventurers” Playlist Got an AMD GPU and a thirst for local AI? This playlist is for you. Unleash the power of on-device speech-to-text!

• Episode #2: Local STT For AMD GPU Owners

Episode #14, “AGI’s Crossroads: Are LLMs a “Dead End” to True AI?”, stands out this week. While many episodes focused on the practical applications of current AI, this episode tackles the big, philosophical question of whether the current path of LLMs is actually leading us towards true AGI, or if we’re headed down a dead end. It’s a provocative question that forces us to think critically about the future of AI.

By the Numbers:

• Number of episodes published: 14
• Top 3 categories/themes: Speech & Audio, AI Core, Fine-Tuning
• Notable patterns: A serious ASR focus this week! Nearly half the episodes tackled some aspect of Speech & Audio.