You are right with the whole toc but in their proceedings systems you can filter by topic and then it feels similar in scale to aiide. So yes, the entire catalogue would kill me and I also do not care about a lot of topics and still get interesting variety after the filter.

I had not thought about this, for me it is right now DIS,CHIPlay,CHI,TEI as I find lots of inspirational papers on the fringes.

For some models this could be the shared user profile though, so the extra information the user does not enter into the prompt to create the soft prompt.

Yes that is true but the point is that for a project I am looking into, the outputs and the activations are available. I am also not entirely sure how well it works because I did not dig in sufficiently but if you get activations you should be able to reconstruct the input, not the training data.

It might have or need 4 of those ;) but what else does it entail?

The key issue for me would be what you define as intelligent? If you mean perceived behaviour from a human observer, I think that is possible. Additionally, we attribute a lot to embodiment so we would need a cat-ish robot body as well.

Not necessarily, I have heard that also from musicians and performers, for music I think your argument holds especially considering the high degree of Automation in Mainstream music for performance it feels different as genAI can't easily replace a performance.

The only coffee worth getting :)

Hi Joris, I hope you had a good vacation. I can't dm you somehow, would be great if you had some time to chat.

Spyros Samothrakis from Essex is presenting this

How to think about games as a way of approaching AGI,... maybe?