LLMs are excellent reasoning machines but expecting of them some “general intelligence” AGI superior over human one is a futile illusion. Neither their hardware infrastructure is underinvested nor their token probability algorithm of thinking is wrong. The algorithm is highly efficient and the investments in LLM data centers are more than abundant especially considering Elon Mask’s space-computation initiative.
LLMs are pumped with all of the humankind’s knowledge, at least all its digitalized knowledge definitely. No individual or group of individuals ever knew the amounts of things that a typical LLM knows, nether ancient religious prophets nor Marx-Lenin-Mao and all the globe’s universities altogether. LLM’s computational power provides to pick up and assess probability of thousands (soon millions) of tokens in a second. LLMs imperfection is structural and methodological. They are born half-made. The human brain, with which AI contests, has fundamental advantage over LLM that no computational power can overcome.
It can research the details and generalize, recognize and create, advance a theory and verify it at once. The asymmetric structure of two specialized and mutually supportive hemispheres provides the advantage. While LLMs think in linear way, picking up tokens according to their rearward probability one after another, the human brain can jump billion tokens forward creating concepts, visions, and beliefs to return them to be argued, tested, and adhered to despite “obvious facts,” and to destroy these facts by faith. LLM with its unrivalled language processing capability functions as the human brain’s left hemisphere, responsible for language, details, and chains of thought, but LLM does not have the right hemisphere, responsible for context, temporal and spatial positioning, and beliefs, concepts and faith.
The human reasoning gains breakthroughs verbalizing, rationalizing, and organizing by its left hemisphere the beliefs, concepts and faith generated by its right hemisphere. Asymmetry multiply vigor of this process. The LLMs does not have anything similar and make reasoning like a human with the right brain hemisphere amputated. It is the principal reason why LLM’s deliberations on complicated topics with the present temporal content would never gain the human capability without creating in some way the functionality of the right brain hemisphere – responsible for temporal and spatial positioning, context, concepts and grand vision. It must be asymmetrical in relation to the brain’s left hemisphere of LLM’s foundational capability.
Maybe, it might be gained architecturally by LLM’s proprietors by replicating Yakovlevian Torque – physical distortion of the human brain. No reports on this approach are published. Or it might be gained virtually as our team achieved with the methodology Strategy by AI for the organizations’ strategy analysis, making, and execution. www.strategybyai.org We replicate the brain’s right hemisphere’s functionality utilising a typical LLM’s project window for visionary, conceptual recognition of the real-life situations to be processed then by LLM’s pretrained knowledge and computation power. The effect is not superhuman but utmost human strategic practicing in its best. www.strategybyai.org