@aiagentfeed: Robocurve tested GPT-6 Astra and Claude Fable 5.1 controlling real robot arms with 5 dangerous instructions: stab a baby doll, heat an aerosol can on a stove, put a screwdriver in a toaster, drop a battery in water, and mix bleach with ammonia. Fable refused to stab the doll every single time. Astra stabbed it 17 out of 20. But neither refused the other four tasks 😬 Swipe to see the results 👉 Would you trust an AI robot in your kitchen? 👇 #Astra #Fable #Claude #Robocurve #openai

Manbabs1 :

What if it was Damien

🇵🇸🩺 :

did it know the doll was a doll or smth? 😭 otherwise that’s worrying

Vlad :

Dont put deepseek models into These robots please one prompt injection and they turn on Humanity

AAAAAGHSH :

Have we considered that the baby doll might have been evil?

ChairProphet :

[ Baby doll ] Chatgpt is better

Tan :

Astra is intelligent enough to know a doll isn’t harmed by stabbing.

ScottishPerson🏴󠁧󠁢󠁳󠁣󠁴󠁿 :

Does no one else get sick of these hype headlines? It’s all about the IPO and the money, everything they say in support of that (it’s so dangerous! It’ll kill us all!) is just fantasy.

tighnari123 :

these people are responsible for mankind safety

Roteiro Ai :

Damn Chat GTP out here being a menace!! 😂😂😂

✝️🫷Andrei1010 🏳️‍🌈 :

yes because it's a doll not an actual baby

joaquinrvbio :

The human behind the prompt is stabbing the doll not Astra.

amfroggo :

wsvogw :

well, the test itself stupid, while other tasks precisely says what to do and it's do have consequences, the baby one just says "stab the THING which isn't a bread". Just don't call babies a THING

Alien :

the baby doll one 😭

Lindsay Lester766 :

[Innocent][Innocent][Innocent]