3 ms·
It's hardly surprising that a LLM trained on thousands of texts relating to these objects would give more interesting answers than the average human. How is th
by imiric 3y ago
It's hardly surprising that a LLM trained on thousands of texts relating to these objects would give more interesting answers than the average human.
How is this a sign of creative thinking?
- mortureb 3y agoBecause that’s what humans do as well.
- imiric 3y agoSo humans were allowed to consult Wikipedia?
- thefurdrake 3y agoIs this even a real question? Yes. Yes they are. If a human has looked at wikipedia, that information is in their mind and available to be used as creative fodder. That's how creativity works; you experience something, it creates a memory, and you draw on that later with other memories to confabulate new information. I suppose it's possible none of the tested individuals have ever gone on Wikipedia, though. I could be wrong!
- imiric 3y agoI meant during the study. The LLM is capable of instantly recollecting all the information it was trained on when prompted with a question. Human memory and processing are far less capable. So the results of this test depended on a) whether humans were "trained on" relevant information, b) their capacity to recollect it, and c) their capacity to produce new information, i.e. creativity. The only fair comparison would've been to allow humans to consult the same information the LLM was trained on when asked the question, give them time to process it, and come up with an answer. Otherwise, this "study" is only proving that machines are better at processing data than humans, which is undisputed, and not anything related to creativity itself.
- flangola7 3y agoHumans recollect on what they have been trained on as well. I'm not seeing the tangible difference.
- thefurdrake 3y agoAh. I suppose that is fair. Maybe. Almost. ... I still don't know though. It feels like humans still have an out-of-band advantage in how we /remember/ data for creativity. Maybe I'm just not prepared to understand how we can quantify "creativity" between two different entities with vastly-different capabilities yet.
- blovescoffee 3y agoProcessing/parsing/digesting data so that it can later be used is crucial to creativity, else the creator is just a blank slate.
- jug 3y agoYes but an AI will have the collective knowledge at their fingertips. I think that’s a main difference here — humanity as a whole can progress very far but singular humans can’t be compared to an AI.
- alexfromapex 3y agoIt’s not, unless someone can prove that an AI is having an original thought. A thought which isn’t a combination of other non original thoughts.
- david-gpu 3y agoCan you prove that any person is having an original thought; a thought which isn't a combination of other non-original thoughts? Or does that standard not apply to humans?
- alexfromapex 3y agoYes, you can. AI can mimic some behavior well by using simple regression. AI hasn’t shown the theory of mind necessary to have a legitimate thought which is the first barrier before then having an original thought which would deviate from the norm and display imagination, critical thinking, and synthesis of diverse information.
- NoZebra120vClip 3y agoLLMs are ideal examples of derivative logic, and their output is always a derivative work. So for humans, the line between "creative work" and "derivative work" can be difficult to discern, and sometimes even needs to be decided legally in a courtroom. The unique nature of the LLM, though, is that they can turn out derivative works at a prodigious rate, derived from an enormous corpus of work, and the sheer volumes of these inputs and outputs causes humans to just throw up our hands and decide that the LLM is a creative entity after all.