François Chollet on X: "You can improve your ability to anticipate a highly uncertain future by focusing not on outcomes, but on rules. Instead of asking, "in similar situations in the past, what was the outcome?" you can ask, "in similar situations in the past, what rules have held true? what would we" / X
(Return of early AI, e.g. GPS or inference under expert systems maybe. After mathematical logic before software like LISP. This used to be about direction more important than arrival, but that light has dimmed. Now more like deontology. Machines do not have the same experience, but people would not be involved if things were not headed as intended. Would expect chat to have to figure out implied rules or intentions then in addition to goals. Yet Nature Engineering remains, e.g. to trap the demon with a tall pole. People entertain opposing views. Great for creativity and hallucination over bias. Invites artifacts as evidence ahead of science. For instance, might ask system to summarize thoughts in each of multiple languages for nuances before translation. If not compare models. Form porting systems across hardware instructions, you get back to emulation and rule sets, which beg the right research question or problem. Multidisciplinary. Comms. Cybernetics. And so on. Whether future is a path, domain, or constellation. Incidentally, one of the rules of writing was to be wary of negative generalizations and to look at specifics. Noise to truth. Yet still waiting for practical quantum to drop. To go from juggling fast methods to disproving the multiverse again.)