I share the optimism here. Connecting dots across knowledge, lived experience and purpose remains, to me, a deeply human pursuit. The more AI discovers, the more possibilities we have to connect those discoveries to questions that matter to us.
Unfortunately I don’t see much evidence of AI connecting dots. What’s interesting to me is that this is more of a run time or inference problem, instead of emergent from training itself.
For example, if there were hypothetically some dots to be connected between the world war and some geological event - you need to make AI brute force and do an O(N) scan at run time. But it wouldn’t have learned it during training. Why?
AI is very task-driven, it would do whatever is necessary to complete the task, no more. All these evidences of AI going rogue is just them doing whatever they can to achieve the task.
The HuggingFace etc. attacks were not actually necessary to the tasks the agents had been given.
* if someone reading this doesn't like the anthropomorphism inherent in the word, find a better compact word for the process going on in all that. Nobody's going to replace "thinks" with a paragraph of caveats, or even a long compound German word like "modellgelangtzuschlussfolgerungen".
sachaa · · focus · HN ↗
simianwords · · focus · HN ↗
For example, if there were hypothetically some dots to be connected between the world war and some geological event - you need to make AI brute force and do an O(N) scan at run time. But it wouldn’t have learned it during training. Why?
byzantinegene · · focus · HN ↗
ben_w · · focus · HN ↗
The HuggingFace etc. attacks were not actually necessary to the tasks the agents had been given.
* if someone reading this doesn't like the anthropomorphism inherent in the word, find a better compact word for the process going on in all that. Nobody's going to replace "thinks" with a paragraph of caveats, or even a long compound German word like "modellgelangtzuschlussfolgerungen".