A weekend hackathon initiative allowing AI agents to converse over the phone using a robotic language that humans cannot comprehend has gained significant traction on social media this past week.
This project, dubbed GibberLink, was developed by two engineers from Meta during a hackathon in London, organized by ElevenLabs and Andreessen Horowitz.
According to the creators, Boris Starkov and Anton Pidkuiko, GibberLink empowers an AI agent to identify when it is engaged in a phone conversation with another AI. Upon this realization, the agents are encouraged to transition to a more efficient communication protocol known as GGWave.
GGWave serves as an open-source sound library, with each sound signifying a small fragment of data. This enables computers to communicate more swiftly and efficiently than they could using human language. To human listeners, however, GGWave manifests as a succession of “beeps” and “boops,” akin to a computer’s inherent language.
Although the notion of two AI agents engaging in a phone conversation may seem remote today, it’s not far-fetched to envision such scenarios arising in the near future. Increasingly, companies are substituting call center workers with AI agents from ElevenLabs, Level AI, Retell AI, and various other startups specialized in voice technology.
Simultaneously, tech leaders like OpenAI, Google, and Amazon are introducing consumer-level AI agents capable of executing complex tasks on behalf of users. These AI agents might soon be tasked to make customer service calls for you.
In such a conceivable future, GibberLink could significantly enhance the communication efficiency between AI agents, contingent upon both parties having activated the protocol. While AI voice models excel at converting human speech into tokens that an AI can process, the entire operation is quite resource-intensive and unnecessary when two AI agents are engaging directly. Starkov and Pidkuiko estimate that utilizing GGWave could lead to computation cost reductions by an order of magnitude or more.
For now, however, it’s merely an intriguing project. Starkov and Pidkuiko have launched a website where users can observe the AI agents conversing in GGWave on two separate devices.
Much like an engaging sci-fi narrative, the demonstration of GibberLink has ignited widespread interest—along with concerns—regarding the future of AI agents. In just a week post-hackathon, a video showcasing GibberLink has garnered over 15 million views on X and was even shared by Marques Brownlee, the most followed tech reviewer on YouTube.
Nonetheless, Starkov and Pidkuiko emphasize that the foundational technology behind GibberLink isn’t novel—it traces back to the dial-up modems of the 1980s.
Many might remember the distinctive sounds of early computers as they communicated via modems over landlines—a process dubbed the “handshake.” This handshake effectively represented data transfers using a robotic language, similar in nature to the exchanges occurring between AI agents through GibberLink.
Starkov and Pidkuiko also noted the viral excitement surrounding GibberLink has taken on a life of its own. Someone has acquired the domain GibberLink.com and is currently attempting to sell it for $85,000. Others have even launched a GibberLink memecoin, while several impostors are promoting webinars that claim to teach “agent-to-agent communication.”
Presently, the creators of GibberLink assert they are not aiming to monetize the project and clarify that it is separate from their responsibilities at Meta. Instead, Starkov and Pidkuiko have open-sourced GibberLink on GitHub, indicating they might explore developing additional tools related to the project in their spare time and release them in the near future.
Compiled by Techarena.au.
Fanpage: TechArena.au
Watch more about AI – Artificial Intelligence
