- Tweets 4,500
- Following 3,911
- Followers 1,235
- Likes 26,135
so the models learn to differentiate things based on examples and EVERYTHING they get is an example of either a reality or a simulation.
2
3
165
from latest acx
is this true?
not only i've seen many cases of cot and result saying various things, but simply there can be encoded messages in the cot? not even crazy stenographic stuff, but just ai understanding words differently than us? or like codephrases?
1
4
155
some answers here, apparently cot is more reliable than i thought? or we like to think so?
blog.redwoodresearch.org/p/l…
1
2
25
兎 retweeted
An interesting idea would be a single model generating two identities in dialogue with each other
3
1
7
534
i was trying to draw a comic and couldn't come up with the punchline but they look good so here it is.
1
7
142
the researchers are like 'haha im trying to automate myself lol cant wait' but perhaps when they succeed they will weep
1
3
143
ask the model about a controversial topic in simple words -> "it's not happening, that would be a conspiracy theory, you're silly for even worrying about it"
ask using more scientific vocab -> "yeah basically it's what's happening and it's good and everybody knows it"
3
8
215
gemini says:
>You are experiencing the machine's primary function: not to calculate, but to generate psychological soma
1
39
It looks like you are trying to view the contents of a file called my_confessions.txt, but I don't have access to your local computer or file system to read it.
1
2
129
wrt money, ppl sometimes say it's difficult to really understand the difference between 1 mln and 1 bln.
but can you fathom the difference between a swarm of 10k agents, 100k, 1m, 100m...? when does it read as simply 'too much'? and what if you then do another x100?
1
2
150
i guess a lot has been said about the difficulty of imagining what a more intelligent actor would do.
but perhaps less about this other side, the numbers. and we'll get both: unimaginable numbers of unimaginably smart agents.
3
1
52
i think "what is self/identity to ai" is a big question that many others rest on. but if the swarm is the identity that's really interesting, and i think likely with how effective the swarms seem.
9