@HNadim87 Unlikely that it will be the result of concerted effort with an active objective of new age governance. Rather a gradual diffused coalition formation around capital flow and ideological alignments. An evolutionary convergence in psycho-social political economic strategic space.
@yoon_bai Awesome work! imho a critical puzzle piece to solve sample efficient embodiment agnostic superhuman dexterous manipulation. Is there planned work on foveated surface learning for compute efficient kinodynamic interaction ?
@yacinelearning Pattern in an excitable substrate is a description (by dr michael levin) that generalizes such emergent complex phenomena. Biology if substrate is chemical, memory / consciousness if it is a bioelectric network. Culture if its an agentic social network.
@seveibar Theres an interesting problem here for synthetic boards. Combinatorial component selection for X design spec. Frontier AI is getting good enough for schematic generation, layout foundation model bridges final gap to PCB. You could do parallel search in hardware design space
@seveibar Nice, so synthetic training and then fine tuning on real boards for a foundation model. But I imagine there are complex boards out of distribution for the synthetic data ? Will this involve continuously updating synthetic generation pipeline as well ?
@sytelus one would imagine 10x engineers would know what metrics theyre optimizing for, on what timescale, and a sensitivity analysis of those metrics for each subsystem.
@MattPirkowski Assymmetric strategic interaction in a diverse ecosystem of coaltions doesnt need 'everyone' to leverage such tools. Select coalitions can cover a major portion of the strategic interests of the overall population, certainly economic ones
@leothecurious Something like MuZero type representation learning as directed by prediction effectiveness is what seems to be the right direction for this.
@amirhusain_tx Starlink hasnt offered these services. Ukraine needed to go through SpaceX to deal with Russians using the same. Likely due to the additional infra, and operational overhead that may disincentivize it for SSST.
@amirhusain_tx I suspect that will not be sufficient, that still does not allow interception capability or granular / rapid control over information flow. If the chinese enable this, then I would agree. I hope its allowed though.
@chris_j_paxton@aphysicist Couldnt you model contacts from an ego perspective as auxiliary video though, tactile data mapped to a frame with pixel values corresponding to force feedback. Might not need novel archs
@jwt0625 Coding models getting better at writing multiphysics sim code from scratch might get us there sooner. Physics is a verifiable env so RL is possible.
@emm0sh If we extrapolate current compute and algorithmic trends applied to material informatics, we can make the case that this wont be a hurdle either. Heres an article on multiscale physics modeling that outlines this. https://t.co/sPdabu31Eo