35 THOUGHTS ABOUT AGI AND 1 ABOUT GPT-5

Summary from the TLDR newsletter

GPT-5's release felt uneventful. It is still not Artificial General Intelligence (AGI), and it still gets things wrong - the model is supposed to be able to decide for itself whether it needs to use its thinking mode for harder problems, but many reports show that it often gets this wrong. Large language models perform well with crystallized knowledge, which masks their shortcomings with fluid intelligence. AGI has the potential to change how we do everything very soon after it is achieved, but we still have to figure out how to get there.