Just fyi, he saved my life. I had a heartattack in a burning building and he rescued me to savety and reanimated me. All on his own. I owe him big time. Hes the best
Idk I just think itβs mad weird that this wasnβt edited out at least? The whole bathing scene is just weird af cause the internet is just such a scary place for children and they should be more concerned and considerate of the childβs savety and just not post that kid NAKED
Have u heard of SAns abducting business men to highjack buildings and businesses in foreign countries? Have u heard of SAns run ppl into government places for savety in foreign country? Ur noise is a chamber noise only to yall. Ask yourselves why
Have u heard of SAns abducting business men to highjack buildings and businesses in foreign countries? Have u heard of SAns run ppl into government places for savety in foreign country? Ur noise is a chamber noise only to yall. Ask yourselves why
Have u heard of SAns abducting business men to highjack buildings and businesses in foreign countries? Have u heard of SAns run ppl into government places for savety in foreign country? Ur noise is a chamber noise only to yall. Ask yourselves why
macam macam bekal
mahasiswa rakyat: makanan darurat, savety kit, pakaian, dan poster protes
aparat: full armor, gas air mata, senjata, pasukan 6000
bro, what the hell..
NEW: malware developers added nuclear & biological weapons text to to their spyware.
Goal? To trigger LLM safety refusals... so that their spyware wouldn't be analyzed by an AI security scanner.
Cleanest practical example I can think of for why over-indexing on first order safety alignment is risky.
When closed (and open) models ship with aggressive refusals, they will be sprinkled with second-order blindspots that attackers will discover...and exploit.
We are only in the earliest days of attackers leveraging these features, and it wouldn't surprise me if users systems that need to handle complex cybersecurity issues demand that models be less safety-blunted.
In the weeds: @SocketSecurity's post also shows why intention matters in how you design a malware analysis pipeline to avoid prompt manipulation.
H/T to colleagues that shared this with me socket.dev/blog/mini-shai-huβ¦