3 ms·
Reasoning how? LLMs are statistical models, not brains.
by spondylosaurus 3y ago
Reasoning how? LLMs are statistical models, not brains.
- Energiesustaine 3y agoHow are humans reasoning? Through a logical or statistical internal model. So if you make your model through slow thinking and discovery chatgpt just builds it to a billion words. The result is the same
- spondylosaurus 3y agoThe result is certainly not the same. A LLM (that hasn't been trained on any data regarding this topic) is unlikely to demonstrate the same bouba/kiki effect seen among humans. https://en.wikipedia.org/wiki/Bouba/kiki_effect https://en.wikipedia.org/wiki/Bouba/kiki_effect
- maxdoop 3y agoHow is this relevant, though? People continue to claim LLMs are just parrots. But then there is never a solid rebuttal to how humans are different.
- Energiesustaine 3y agoThat's a very far fetched argument.
- jstx1 3y agoReasoning in behavior - you ask it to explain things and it does; it gives reasons, motivations, nuance, explains connections, clarifies points. That's reasoning.
- spondylosaurus 3y agoThere's a pretty significant gap between being able to give reasons and being able to actually reason (as a verb). The latter requires complex cognitive faculties not present in a mathematical model of the English language.
- jstx1 3y agoLLMs are tool, the important question to ask of a tool is "is this useful", not "does it work like my brain does". I don't see why you care about the internals, and not about the actual outputs.
- spondylosaurus 3y agoThis comment chain wasn't about outputs, it was specifically about whether or not LLMs "reason"—precisely the internals. I feel like you're just moving the goalposts now to deflect from the original question.
- jstx1 3y ago> This comment chain wasn't about outputs My point was that it should be.
- rowanG077 3y agoConsidering we can't even define what "complex cognitive faculties" are you are just talking up air castles at this point.
- cjbprime 3y ago> The latter requires complex cognitive faculties not present in a mathematical model of the English language. The training is on images and text, and the output is text. But nothing constrains the model in the middle of those two ends to only model based on language features, as opposed to world features discoverable from the language and images.