🧠 Recently, a paper titled "Sparks of Artificial General Intelligence: Early experiments with GPT-4" was published, in which scientists claim that GPT-4, the latest model from OpenAI, exhibits the beginnings of artificial general intelligence (AGI). Surely, you haven't missed it.
In the article, the authors demonstrated that GPT-4 is capable of solving complex tasks across mathematics, programming, medicine, law, psychology, and much more, without the need for special query tuning.
Despite the impressive results, I am quite skeptical about the conclusions and reasoning. In my opinion, claims like "it can draw images even though it has never been trained on them, only on texts" are speculative, as there is a wealth of text on the internet that describes image generation (exactly what the model offers in tests). Experiments with stacking objects also do not excite me, as this is quite in the style of synthesizing similar tasks.
What do you think about this?
Paper: https://arxiv.org/abs/2303.12712
Brief video overview of the publication: https://www.youtube.com/watch?v=wHiOKDlA8Ac
Lecture at MIT based on the paper: https://www.youtube.com/watch?v=qbIk7-JPB2c
Analysis of the paper on lesswrong.com: https://www.lesswrong.com/posts/BzfMaGKjAEhLwYC9J/an-overview-of-sparks-of-artificial-general-intelligence
