The conversation centers on a recent blog post about recurrent neural networks and their surprising capabilities. Practitioners collectively praise the educational value of the post and cite similar historical work, noting that RNNs can generate coherent text, code snippets, and even music, which many see as a clear step beyond simple n‑gram models. A recurring point of agreement is that the models remain opaque and lack a solid theoretical foundation, a concern echoed across several comments. Dissent appears in debates over whether the results are truly unreasonable — some commenters argue that comparable outputs can be achieved with well‑tuned Markov chains — while others question whether RNNs ever capture meaning or pose broader societal risks. The exchange also splits on the long‑term implications for programmers, with some viewing RNNs as a looming threat to coding jobs and others treating them as merely an interesting research toy.