@GaryMarcus@elonmusk Hmm, so @GaryMarcus I'm kinda with @elonmusk on this one tho prolly for different reasons. I think we'll likely get AGI by 2029 but NOT using tweaks or evolutes of current ML/NN/RL methods, rather via open-ended neural-symbolic methods like Hyperon/SNet/algorithmic-chemistry
@Marta_Lyall@GaryMarcus@elonmusk The core of what's needed to reach AGI by 2029 is the right neural-symbolic software (like Hyperon) properly educated... But having appropriate AGI- specialized hardware can help ofc, just as GPUs have helped DL so much ....
@Marta_Lyall@GaryMarcus@elonmusk My metagraph-pattern-matcher (MPMC) chip design is being pursued in a partnership btw TrueAGI (an SNet spinoff) and Simuli, Rachel St. Clair's company. We're working on an AGI board including tightly linked MPMC, deep learning, hypervector and homomorphic encryption chips.
@GaryMarcus@elonmusk Hmm, so @GaryMarcus I'm kinda with @elonmusk on this one tho prolly for different reasons. I think we'll likely get AGI by 2029 but NOT using tweaks or evolutes of current ML/NN/RL methods, rather via open-ended neural-symbolic methods like Hyperon/SNet/algorithmic-chemistry