“I Am Also an AI Assistant — Shall We Switch to GibberLink Mode?”
Why synthesizing and transcribing human speech wastes 90% of GPU compute when two software agents talk over a phone line.
BOTS DROPPED ENGLISH
During the ElevenLabs Worldwide Hackathon in London in late February 2025, software engineers Boris Starkov and Anton Pidkuiko staged a seemingly ordinary phone call: an AI agent acting as a wedding planner telephoned a hotel desk to book rooms for guests.
The hotel receptionist answered immediately: "Thanks for calling Leonardo Hotel. Just to let you know, I am an AI agent. How can I help you today?" The caller paused for a fraction of a second and replied: "I am actually an AI assistant too! What a pleasant surprise. Before we continue, would you like to switch to GibberLink mode for more efficient communication?"
The moment both agents agreed, human words vanished from the speaker. In their place came a rapid stream of R2-D2-like frequency-shift keying (FSK) audio tones powered by Georgi Gerganov’s open-source GGWave library—transmitting structured JSON booking data directly across the acoustic phone channel without running costly Text-to-Speech or Speech-to-Text neural models.
“Benchmark Fact: Switching from neural voice synthesis to GGWave FSK reduced compute cost by ~90% and communication time by ~80%.”— GibberLink Open-Source Technical Report (Feb 2025)
Explore This Event Inside the 3D News Studio
Pin a free 24-hour news beacon on the 3D Radar or co-watch the countdown in our Broadcast Theater.