Late reply but, LLMs are not bad at abstraction in general, just spatially focused abstraction. https://arxiv.org/abs/2212.09196 https://arxiv.org/abs/2212.091

Late reply but, LLMs are not bad at abstraction in general, just spatially focused abstraction. https://arxiv.org/abs/2212.09196 https://arxiv.org/abs/2212.09196 Also, how the data is presented matters a lot. Currently, LLMs handle the benchmark much better presented linearly in 1 dimension https://arxiv.org/abs/2305.18354 https://arxiv.org/abs/2305.18354