Andy Halliday

And they did so by using the existing models, including 5.2 and Gemini 3.0 and Opus 4.5, and working then together in a multi-agent kind of system to...

aced the test that i when i say ace i mean they ended up at state of the art yeah and way beyond the level of state of the art of any one model that had been trained you know and assisted in training you know you'd be capable of doing that

Now, DeepMind's approach to this and what gave rise to a lot of concern and commentary is that DeepMind's thesis is that this is going to happen, and that's where we're going to see AGI first.

And that means that alignment and safety is no longer the responsibility of the one model developer who's creating something that could be AGI.

Right.

So now we have to completely rethink and deep mind, of course, did deep thinking with their deep mind on this question.

And they address the entire sort of ecosystem of systems and regulations and protocols and support that would be required to keep.

artificial general intelligence in these multi-agent systems in check and in alignment with human interests.

That is very interesting, but I'm more interested in, wait a second, are we already seeing AGI in systems like the one that Poetic demonstrated?

And will we now see a dramatic acceleration towards AGI?

And that AGI may in fact come

in this year as a result of improved orchestration, improved harnesses for multi-agent actions.

And what's happening there behind the scenes is, you know, there's protocols emerging that are making it easier and easier for agents to work with each other and to consult with other

even remote agent systems, agent to agent from Google and so on.

And somebody is going to build a very, very smart AI intelligence that isn't confined to working with a single model.

It's not like Prometheus.

In the Max Tegmark Life 3.0 book, it's not going to be one model that achieves this and becomes superior to all others.

It's going to be everybody out there figuring out how to use all the now interconnected tools in AI to achieve AGI with a master orchestrator on their desktop working with agents that are distributed around the world.

Well, let me ask this question of you both.

Is there really a material and substantial identifiable difference between cloud code, anti-gravity, cloud codecs, and a host of other coding-focused multi-agent systems out there?