Transformers are the final architecture needed for AGI

Created by PseudonymousUser on 2023-01-18; known on 2099-01-01

• PseudonymousUser estimated 85% on 2023-01-18
• PseudonymousUser said “Prediction is agnostic to training curriculum used, data availability issues with AGI, training data modalities, use of auxiliary architectures like convnets for vision, etc. If 2099 occurs before AGI, prediction is marked as false. ” on 2023-01-18
• PseudonymousUser said “If something comes along and replaces the transformer the way transformers replaced LSTMs, and that something proves necessary for AGI, then the prediction would also be false. ” on 2023-01-18
• chemotaxis101 estimated 72% on 2023-01-19
• dmwit estimated 8% and said “I give “no transformer at all” about 80-85% chance (e.g. “breakthrough in asynchronous architectures” seems fairly likely). When “no transformer” fails, “yes transformers, but at least one nontrivial non-transformer component” seems quite likely.” on 2023-01-20
• sty.silver estimated 10% on 2023-01-23
• Baeboo estimated 25% on 2023-01-26