Is this the true power of AI? Why am I paying for this shit?
Because you've got to make the Monthursday deadline.
>>106495725>he doesn't know about MonwednesdayKek. Memory hole successful.
what's the problem? Extra enum values do no harm, the value you need got generated
>>106495725>>106495748For me it's Monfriday.
>>106495725>this nigga works on Mowednesdaywhat a wage cuck
>>106495725I don't mind a little occasional vibe coding but those autocompletes absolutely get in the way when I do want to write code.
>>106495751What happens on Monwednesday?
>>106495927For me it's monsaturday, because we just shorten that to monsterday.
>>106495725>>106495748>>106495751>>106495811>>106495927>>106495941>>106495953>>106495965>>106496173The real answer is that someone out there, probably Eastern Europe, really went ahead and commited Monwednesday to some obscure github repository and now it's part of Microsoft's AI model forever.
>>106496239Nah, it's hallucinating names because it doesn't know what comes after Wednesday.That's the big problem with AI: if it doesn't know something rather than owning up to that fact it will either talk a lot around it or say something wrong.AI is really designed to fool people into thinking it's more intelligent than it is.
>>106496322>AI is really designed to fool people into thinking it's more intelligent than it is.That's everyone in my department WTF
I hate Montuesdays
>>106496528Lasagna.
>>106495751AI casually revealing the thremboth day of the week
>>106496239>probably Eastern Europedas raycishttps://github.com/Duongnguyen040902/edunexus/blob/master/source-base/sep.backend.v1/Common/Enums/DayOfWeek.cs#L8
>>106495725>Why am I payingYou aren't. You're using some dogshit model to further your anti-AI agenda on a Vietnamese basket weaving website as if your opinion matters at all.
>>106496680>it's realholy fucking shit>>106496322you got fucking destroyed mate, my intuition was right
>>106496322Is there a technical reason why LLMs can't say they don't know? E.g. would it interfere with the loss function or gradient descent if you tried to add a don't know option?Or is it just a design choice to always give some answer to fake competency?
>>106499226I think bc its already trained you cantr eally check how few data were used to create the "synapses" all you can do is go from input to output. Maybe if you inserted some sort of confidence meta data during training? I dunno i forgot how it works exactly
>>106499226Rarely do you ever see a hard question on the internet answered with "I don't know". It is simply not in the training set and even if it were, like if humans did such a thing habitually, this would expose the inherent weakness of LLMs.
