<J24k>
GPT4 got that question right - and i wonder if we are not also just a language model - Ü
<teepee>
but not due to intelligence, understanding the meanding of it, only by now people have asked it a lot and statistics shifted
<J24k>
teepee but "Her" is about 10yr old and if GPT4o is able to do that - i am a little impressed
<teepee>
I'm more scared about all the potential misuse that will be explored soon
<J24k>
we know that our brain is just a big probability guesser .. and the difference from apes to human is just more neurons .. there are no special new neurons
<teepee>
so far that's a guess too
<teepee>
the main problem I see is that the technology is not used to learn and understand both human brain and AI
<J24k>
science nowadays finding more and more clues that humans are not so special at all .. yes there are some exceptional but most of them are mundane
sadome has joined #openscad
sadome has joined #openscad
sadome has quit [Excess Flood]
<InPhase>
https://bpa.st/MOOA :( I thought the offline models I had would see through that one, but they both tripped over it hard.
<InPhase>
Those were my top models for handling trickier situations.
<teepee>
ouchies
<J24k>
I mean this questions are sort of trick questions as children often get these wrong because of the contradiction that something that is heavier (denser) can be lighter
<InPhase>
J24k: Yes. Although we expect slightly better than children level processing from the top models.
<J24k>
i see the problem that they are unaware of their limits and claim to be all knowing .. and some humans believe them (and politicians)
<InPhase>
Son of a... I thought this was just pattern matching tripping it up on feathers and lead being a classic problem with too much training data of bad responses. So I changed the parameters: https://bpa.st/IPZA
<InPhase>
Watching this generate in real time, I was like "Yes! That!" until it got toward the end...
<J24k>
if humans get this wrong .. so does AI - i mean all that trainings data is mostly internet trolls
<teepee>
good, so it's at least stolen stuff + troll garbage
<InPhase>
J24k: That is a big part of it.
<teepee>
after all, that's exactly how we learn too.... wait, we don't (yet?)
<J24k>
i mean we also use solutions that we don't always understand .. and just hope they are true and working
<teepee>
depending on the topic, I might be ok with that
<teepee>
I'm not when talking for example air plane safety or bridge building
<J24k>
I use my own solutions without understanding them - hoping i understood them when i came up with the solution
<J24k>
Nobody is calculating that by hand - they all hope the computer program is correct
<J24k>
like PI is used by quite a lot more people than people can calculate PI
<teepee>
*using* something you don't unterstand is normal, yes
<teepee>
the people creating that thing usualy do
<J24k>
they know what they do but rely also on other modules and that these are working proper (which is not always the case)
<J24k>
it is not that people would make computer software from the ground without any help
guso78k has quit [Ping timeout: 250 seconds]
L29Ah has quit [Read error: Connection reset by peer]