in thread "Accelerationism": If intelligence is the energetic efficiency of a system, are LLM's getting more intelligent with increasing model size or less intelligent? Seems quite energy-inefficient to me even though there is for sure some axis along which LLM's get more intelligent ... 7mo ago (collapse hidden) log in to judge received 4.0 4.0 If intelligence is t (view hidden) log in to judge received 4.0 4.0 Seems like we have very different intuitions of how efficient LLM's and humans are. Could you flash out your thoughts or maybe even calculations of why you think that is? It sounds to me like what makes LLM's more efficient for you than humans is just the ... 6mo ago (collapse hidden) log in to judge received 1.1 1.1 Seems like we have v (view hidden) log in to judge received 1.1 1.1