$30 billion and the water supply of a small city later
-
@pwinn @mistakenotmy older model with different random word
@pwinn @mistakenotmy All models available under free subscription are able to produce correct answer for different random words
(haven’t tested with Opus 4.7 since I don’t have pro plan account under my hand) -
@pwinn @mistakenotmy All models available under free subscription are able to produce correct answer for different random words
(haven’t tested with Opus 4.7 since I don’t have pro plan account under my hand)@pwinn @mistakenotmy … and completely random (presumably not hardcoded) question, just in case
-
@pwinn @mistakenotmy … and completely random (presumably not hardcoded) question, just in case
@pwinn @mistakenotmy ... and to put some serious nails into the coffin of "LLMs are dumb and can't solve puzzles" take -- here's Hack The Box CTF profile of my Sonnet 4.5/4.6 based AI bot: it can solve insane difficulty tasks and performs on the same level with top 0.5% of human players. Most of these tasks are recent ones so it doesn't have any writeups or solutions in its training data. So yeah: trust no one and conduct your own experiments
-
$30 billion and the water supply of a small city later
@mistakenotmy Sorry, I call cap.
-
@mistakenotmy I got curious and tried adding a system prompt to Claude basically saying "always use tools, don't trust yourself, always verify if possible", and then it got it right. They tell the models to act like this because it drives more engagement when they are confident and answer quickly on things that are deemed "trivial". That makes it somehow worse.
@pontus_k my custom prompt is even simpler. Be terse, don't be afraid of challenging my opinion. I've attached my own results and for both questions it's a one character: "1".
-
$30 billion and the water supply of a small city later
3 years of kindergarden, 8 years of primary school, 3 three years of highschool and three years of university and uncounted media consumed in between and all the resources wasted and the stupid human misspelled "strawberry"
What a colosal failure.
Euthanise this aborted biological. -
@CJPaloma @iwein @mistakenotmy
It's a poorly formulated reply that makes you seem like an AI apologist to me. Especially this part:> The problem only arises when idiots claim a tool can do anything one can dream of. Blame the idiots, not the tools.
Tools & technology have politics. They aren't neutral. AI in particular, regardless of its capability or who wields it, has the politics of surveillance & oppression, environmental pollution & exploitation, labor disruption, and more.
I doubt this is the first time you've seen it characterized as such on the fediverse. So it also reads to me like you're turning a blind eye to those criticisms, many times expressed more eloquently than me.
If you are opposed to AI (genAI and LLMs), consider that you aren't coming off that way currently.
@toolbear all that is ok
-
R relay@relay.publicsquare.global shared this topic
