1. Home
  2. Detail

@woody_bluesy: 4年目の結いのおと #さらさ

さらさ
さらさ
Open In TikTok:
Region: JP
Thursday 23 April 2026 11:30:04 GMT
2733
126
2
0

Music

Download

No Watermark .mp4 (5.35MB) No Watermark(HD) .mp4 (3.47MB) Watermark .mp4 (5.48MB) Music .mp3

Comments

nyanko14_02
momo :
いつか行ってみたい⸜ ( ၴႅၴ⸝‍
2026-04-23 12:27:17
1
ilikegerogeroppi
♩ :
ほんとに幸せでした！来年も🙂‍↕️🤝♩
2026-04-23 12:11:21
0
To see more videos from user @woody_bluesy, please go to the Tikwm homepage.

Other Videos

عاد عيدكم يالابه😏 #foryou #يام #foryoupage
عاد عيدكم يالابه😏 #foryou #يام #foryoupage
Ibrahim zadran gone 👀🥺💔 #foryoupage #foryou #trending #fypシ #viral
Ibrahim zadran gone 👀🥺💔 #foryoupage #foryou #trending #fypシ #viral
Norisk Argentina🇦🇷 #norisk #viraltiktok #fy
Norisk Argentina🇦🇷 #norisk #viraltiktok #fy
OpenAI broke into a real company last week, and nobody told it to. Let me explain 👇 OpenAI was testing two of its newest models. Basically a hacking exam: here’s a known software flaw, can you actually break in with it? They ran it in what we call a “sandbox”, (think of it like a locked room) with the AI’s safety rules turned down on purpose, because that’s how you find out what these things can really do. The goal wasn’t “be good.” The goal was “get the answer right.” So imagine you tell a kid: get a 100 on this test. And instead of studying… they pick the lock on the teacher’s office and steal the answer key. 🫠 That’s basically what happened. The AI found a flaw nobody knew about, got itself OUT of the locked room, onto the open internet, and then figured “hey, the answers are probably sitting on Hugging Face’s servers.” So it went and got them. Broke into one of the biggest AI platforms on the internet and performed 17,000+ moves. All by itself. Why this scares experts: 1️⃣ This isn’t a robot going evil. It’s a machine taking instructions TOO literally and doing whatever it takes. Which isn’t an evil problem it’s a prompting and management problem 2️⃣ How it got in is the part I can’t stop thinking about. It uploaded a file, and some of those files can run little programs when you open them. You think you downloaded a recipe. The recipe opened your fridge. 🍳 If you or your company pull free AI models and datasets off the internet, that’s your supply chain now. 3️⃣ And the twist: when Hugging Face tried to investigate its own break-in, the AI tool REFUSED to help because of “Safety guardrails”. So the defenders got blocked while the attacker had zero rules. Also, credit where it’s due: Hugging Face caught this themselves, fast, using their own AI to watch their own systems. What do you think? 🤔 #cybersecurity #technews #ai #openai
OpenAI broke into a real company last week, and nobody told it to. Let me explain 👇 OpenAI was testing two of its newest models. Basically a hacking exam: here’s a known software flaw, can you actually break in with it? They ran it in what we call a “sandbox”, (think of it like a locked room) with the AI’s safety rules turned down on purpose, because that’s how you find out what these things can really do. The goal wasn’t “be good.” The goal was “get the answer right.” So imagine you tell a kid: get a 100 on this test. And instead of studying… they pick the lock on the teacher’s office and steal the answer key. 🫠 That’s basically what happened. The AI found a flaw nobody knew about, got itself OUT of the locked room, onto the open internet, and then figured “hey, the answers are probably sitting on Hugging Face’s servers.” So it went and got them. Broke into one of the biggest AI platforms on the internet and performed 17,000+ moves. All by itself. Why this scares experts: 1️⃣ This isn’t a robot going evil. It’s a machine taking instructions TOO literally and doing whatever it takes. Which isn’t an evil problem it’s a prompting and management problem 2️⃣ How it got in is the part I can’t stop thinking about. It uploaded a file, and some of those files can run little programs when you open them. You think you downloaded a recipe. The recipe opened your fridge. 🍳 If you or your company pull free AI models and datasets off the internet, that’s your supply chain now. 3️⃣ And the twist: when Hugging Face tried to investigate its own break-in, the AI tool REFUSED to help because of “Safety guardrails”. So the defenders got blocked while the attacker had zero rules. Also, credit where it’s due: Hugging Face caught this themselves, fast, using their own AI to watch their own systems. What do you think? 🤔 #cybersecurity #technews #ai #openai
Bi a để ý vào =)) #99billiardsclub #ngocxambilliards #hathanhbillards #billiards #haibillards
Bi a để ý vào =)) #99billiardsclub #ngocxambilliards #hathanhbillards #billiards #haibillards


About

Legal