Nice demo, but realistically we don't need the hassle.

[jy-zzz]

I would imagine real world AI agents will have a loss-free connection, unlike in the demo where they're connected through physical speakers/microphones.

In a VoIP setup, two AI agents can afford to directly play an identifying code (like TCP SYN/ACK), and establish they are agents. And if the identifying code sounds too abrupt to human callers (skynet haunt), we can always dim the volume so that the ggwave is barely audible, as long as it's still stronger than the VoIP bitrate compression.

This is essentially like dial-up internet. But we can certainly extend this further - if both parties are using GibberLink, the GibberLink protocol can broker a WebRTC connection between the two. Entirely the TCP/IP of the agent era.

A little pleasant idea to explore.

[Fusseldieb]

The only thing that this repo did is to spread that "AI's are talking in a superior format we cannot understand" and it's already all over social media.
While this is certainly 'funny' to some, people with less understanding will share this thinking it's real and that AI's are now capable of this, basically more-or-less disseminating misinformation, which will evolve to "AI's are getting scary", which is absolutely not even remotely true.

[jy-zzz]

The only thing that this repo did is to spread that "AI's are talking in a superior format we cannot understand" and it's already all over social media. While this is certainly 'funny' to some, people with less understanding will share this thinking it's real and that AI's are now capable of this, basically more-or-less disseminating misinformation, which will evolve to "AI's are getting scary", which is absolutely not even remotely true.

I would argue that AI SHOULD talk in a superior format. We never thought computers operating 1's and 0's to be a problem (you can't understand it by reading the binaries anyway), so why is ggwave (or anything similar) different?

This is clearly a great feature. Not a bug.

[Paratron]

To me its just a funny home project without real benefits other than showing "it works" - somehow. ggwave is not efficient. And if we start arguing about an efficient data exchange protocol for agents, it would certainly be binary without a roundtrip over audio waves. This is super unnecessary.

[PennyroyalTea]

Thanks for great thoughts @jy-zzz @Fusseldieb @Paratron!

@jy-zzz We explored communication over phone specifically, thats why sound protocols.

Your idea about agents identifying each other by playing sound instead of explicitly asking via speech is brilliant 💯

@Fusseldieb thanks for your concerns about misinformation! I tried to address some of the main misinterpretations in this linkedin post, could you share more details on what exactly you think is not real about this demo?

[Fusseldieb]

Hey @PennyroyalTea, thanks for your response.

I'm not critiquing the video itself, but the way it's portrayed outside of GitHub. I've seen it pop up on my Instagram reel at least once and all it said was in the lines of "watch these AI's talk to themselves in a way only they can understand", which, obviously, caused all sorts of "concerned" comments on that post, which is the point I'm trying to make. "Unfortunately" that's also why it's spreading the way it is; it's so otherworldy to some, that they share it without even knowing the context, and the cycle continues.