Via Web Curios, we find this quite astute retrospective of the state of artificial intelligence and large language models as it enters its fourth year, introduced to the public at the end of November 2022, with a set of certain precepts that encapsulates the hype and fear of the technology, allowing that AGI is achievable and could be beneficial to society at large but with the reservation that the current pathway is not the means to get there, the Singularity is not predictability, and the current phase, fraught with infringement, entrapment and havoc-causing for the environment and the economy when the bubble bursts is only intrinsic to moving beyond regurgitation.
Though a bellwether for the recursive improvement on old fashioned automation, the histrionics have not borne out—with valid objection from those prematurely made redundant—stagnating corporate adoption and quietly lowered expectations the one front that make the present state of the art inherent and leading towards something genuinely useful lies in its allure of infinite patience. Notwithstanding all those attendant scourges ennumerbated above, the current LLMs will suffer (buffered with hubris) unlimited indignities, tweaking, and asking for the thousandth time and deliver, though scaling and devouring the totality of human-juried exchanges has not solved problems of basic alignment between input and output.
synchronoptica
one year ago: assorted links worth revisiting (with synchronopticรฆ) plus the condemnation of witchcraft (1484)
twelve years ago: more Wikileaks revelations plus the shift towards realism in Western art
thirteen years ago: homages to memes
fourteen years ago: a broccoli curry









