This article https://t.co/1AAHQKKQ02 demonstrates that intelligence is a combinatorial nested structure by showing that cross-domain understanding is limited by within-domain intelligence. In truth, humans rarely exhibit General Intelligence (by this definition) hence fear of AGI
AGI assumes one underlying pattern in the universe of knowledge. This pattern generates 'instance patterns' across diverse domains of knowledge. In this way our theory of knowledge (itself) constrains our ability to see/understand AGI. We require a new knowledge science. #AGI
LLMs are a model of how people talk about the world. The inferences about the world, they make, depend upon the veracity of ‘how people talk about the world’. LLMs may be teaching us more about humans’ perception of the world than they are about AI & how they work. #LLM
Quality of thought scales! The team @anthropicai has found that the likelihood of finding abstract network concepts increases with model scale. {Self awareness may arise from reflecting on the 'next lowest' form of cognition.} https://t.co/NZbM2uA12w
Presenting today at The Council of Graduate Departments of Psychology annual meeting....about the current AI moment - how we got here, what it means & what role Psychology might play in the evolution of AI. #AI#Fasttakeoff#paradigmshift#ML#singularity https://t.co/Xc3rYMFgNm
This argument, between set theorists & pure mathematicians sounds a lot like an argument between xeno & aristotle. https://t.co/sPfuOjEHBP #settheory#maththeory
AI has progressed fastest in domains with data and verifiable results - said by Periodic Labs. Points out the basis of thought - which resides in thinking forward from knowns <break> The best discoveries are the least recognizable. How do we program that? #AIscience
The more you know, the more you recognize how little you know. Interesting Pod from University of Maryland's Psychology Department. https://t.co/PZXlBzYcVP #UMD#Impostersyndrome
I'd like to suggest that we use different language for 'AI Consciousness'. Perhaps we should allow that silicon-based machines may have a qualitatively different experience? If its fundamental substrate is different, wouldn't '#MachineExperience' be so? #AIexperience
Using an AI to predict future scientific discoveries/behavior & matching that against AI discoveries/behavior could predict when the 'fast takeoff' is coming. https://t.co/opYhce3hp2 #fasttakeoff#Genieout#5thgenknowledge
Using graph to distill the abstract core of knowledge: https://t.co/opYhce3hp2 Using the deep knowledge graph AI makes new discoveries which Humans must interpret #AIGraph#AIdiscovers
Passage below from https://t.co/XbN7IOSK37
Also noted: Reasoning can not be boiled down to a single "crux" token. Similar 'localization hypotheses' have been put forward in human learning https://t.co/0FXLXUj2Cp
Perhaps memory is holistic, not reducable #localization#memory