@nextpath.ai: Developer Nate Herk found a very human way to push Al models when they are struggling with a task. Instead of changing the prompt, he simply tells one model that the competing model already managed to do it. GPT-6 Astra gets told that Opus had no problem, while Opus gets told that Astra already figured it out. It is basically the Al version of saying, "Your sibling already did it." Whether it actually improves performance probably depends on the task, but as a prompting trick, it is hilarious. Would you try this on an Al that keeps saying it cannot do something? Source: Nate Herk / X

Kodlach :

With AI no longer being science fiction, why has nobody explored AI going to war with each other. Like it's always humans versus AI, but it's one AI. not multiple. that would be a cool show or book series.

factsnotbullshit :

and then they said "sorry, opus? you're making up names now, the latest claude model is sonnet 3 not that shit" 🤣

TheCCSI.org :

Just like people, reverse psychology and challenging framing activates a motivation loop, a sort of “know it all” impulse.

christopherlgregory :

I already do that. I use AI to critique another AI'S work then tell the 1st what the 2nd said and make it respond back and forth

Sweaty Pants :

I’ve built hundreds of softwares with ai . It is absolutely ridiculous how well this works

CapTehDuck :

the instigator 😭

D3M1G0D :

We did all the time

хамелеон прайм :

Justaddtrail :

Tried gemini, claude, grok and chatgpt to read the braille on the berlin marathon medal, only chatgpt got it right. Claude didn’t even try.

Lucian :

🤣🤣🤣

Juswa :

Are you asian? 🤣🤣🤣

Tristonian :

😁😁😁

Juswa :

🤣🤣🤣

Juswa :

🤣🤣🤣