1.Douglas, R.Hofstadter：Machine translation VS Analogy Thinking

Douglas, R.Hofstadter will explore the topic of human creativity and artificial intelligence through the topic of Machine Translation.

[@chillyfinger]

His grand summary mirrors objections to AI that were voiced in the early days of "AI." It boils down to the fact that a machine cannot "speak" intelligibly because human speech is a web of analogies, ultimately based on human experience that a machine cannot have. What we didn't understand back then was the effect of turning the whole enterprise over to corporations bent on providing commercial products. Now, the general public more or less accepts the idea that we will eventually have a machine that "thinks." Hofstadter is now a lone voice in the wilderness, reminding us of what "thinking" really is.

[@thyhandrevolve]

🤩 Hofstadter 💛

[@chillyfinger]

Hofstadter labors in obscurity as the only real expert on how human language works.

[@slygryphon]

Wow, we have come so far since then (although the video is dated 2022, I think it is actually 2018 based on some of the dates on slides, e.g. at 34:50). Hofstadter is of course right, you can't translate without understanding. The conclusion is then that GPT-4 does understand. Little John was looking for his toy box. Finally he found it. The box was in the pen. John was very happy. 小约翰在找他的玩具盒。最后他找到了。盒子在围栏里。约翰非常高兴。 The roofer came down form the hospital roof carrying a case of shingles. 屋顶工人从医院屋顶下来，携带了一箱屋顶瓦片。 鹿柴 (王维) 空山不见人, 但闻人语响. 返景入深林, 复照青苔上. Deer Enclosure (Wang Wei) In the empty mountains, no one is seen, Yet echoes of human voices are heard. Sunlight enters the deep forest, Again illuminating the green moss. You can verify the translations by asking GPT-4 to explain.

[@swarmaclub7350]

Yes, this video is Hofstadter‘s speech from 2018. Without a doubt, GPT4 has performed impressively in both its comprehension ability and evolution speed. we have been organizing reading club to discusses relationship between GPT and consciousness. If you are interested: pattern.swarma.org/study_group/23

[@fizeek]

Very good video I like a lot

[@fizeek]

One problem: China

[@borntobemild-]

Google and openai actually have far better translation tools that could benefit from Douglas Hofstadters direction with

[@chillyfinger]

True. At least for the examples he provides, context is key. Chat-GPT is good at context because it works with a "knowledge base" behind the scenes, based on huge selections of text. Chat-GPT was available when this lecture was given.

[@stevenjacobs2750]

This talk did not age particularly well! Its pretty funny to see the criticism from Bar-Hillel echoed such few short years ago because what he described as laughable on its face seems to be exactly what we have now.