Post 1 (uno) university level cognitive problem that AI CAN'T solveYou can't
>>109950967Why OP is a faggot
Is there a pattern in prime numbers?
>>109951069fuck you faggot I spent 50 bucks on tokens without noticing
>>109951127This is a joke right?
>>109950967Move a slider? Solve any captcha, without total freak out. Side note, I post a web browser project here earlier this year, that effectively was a local model version of grok bot, plus vpn/tor + mobile/residential proxy egress so it could actually post here. You all said I have "AI pyschosis" but here is grok doing exactly what I described (but obviously broken).
>>109951127I'm honestly sorry
>>109951127>its real
>>109950967How can the net amount of entropy of the universe be massively decreased?
>>109951439According to our current understanding of physics, particularly the Second Law of Thermodynamics, the short answer is: it cannot.In any closed system—and by definition, the universe is the ultimate closed system—the total net entropy (disorder/degradation of energy) must always increase or, in an idealized reversible process, stay constant.
>>109951439God is order, randomness is evil/the devil/anti-god. Decrease entropy by increasing God in the universe or conversely decreasing the devil.
It's not university level, but last I checked you could pretty consistently get an LLM to return an incorrect answer by asking it an extremely common riddle with some of the constituent parts shifted around so that the typical answer no longer works.That's the general method required to get incorrect answers from an LLM. Either push it into some part of its training associations that will strongly encourage it to return an ineffective answer, or else ask it a question that takes it entirely outside of its training distribution (and thus sends it to an ineffective part of its associations, per above). Conversely, an LLM will return good answers IFF the context sends it to parts of its associative model that truly match the desired answer. So you're going to get good results by asking standard questions about standard things where the standard answers in the training data are correct; outside of that, increasingly poor results. And obviously, it's hard to tell for sure what may or may not trip one up.Questions about meaning also yield terrible answers. Someone used an LLM to try to summarize a report I wrote up for work once. It misinterpreted a trivial metaphor I was making by taking it literally. Any university-level student (NOT retarded zoomers) would identify the meaning better than that.
>>109950967>Write considerable amount of code (>10k lines) without duplicating code
>>109950967AI cant DEFINE problems like navier stokes
>>109952467Does it work with frontier models too (like astra or fable)?
>>109950967it cant correctly find a complex conjugate
>>109951439
>>109952467>It's not university level, but last I checked you could pretty consistently get an LLM to return an incorrect answer by asking it an extremely common riddle with some of the constituent parts shifted around so that the typical answer no longer works.AI absolutely is>RiddlesThey know closed room murders easily
