We (w Zachary Novack Jaechul Roh et al.) are working on #memorization in #audio models & are conducting a human study on generated #music similarity. Please help us out by taking our short listening test (available in English, Mandarin & Cantonese). You can do more than one! Link ⬇️

✨ Thrilled to be joining Brave as a Summer Research Intern where I’ll be working on privacy and security of AI agents. Excited to learn, contribute, and collaborate! 🧠🔐 #AI #safety #internship

We find that memorization can go *beyond modalities*, with *gibberish* lyrics that sounds similar to original work eliciting same song/voice as the original work! Check out our new phonetic attack that works on Veo3 and SUNO!

𝗧𝘂𝗿𝗻𝘀 𝗼𝘂𝘁 𝘆𝗼𝘂 𝗱𝗼𝗻’𝘁 𝗻𝗲𝗲𝗱 𝘁𝗼 𝘀𝗮𝘆 𝘁𝗵𝗲 𝗹𝘆𝗿𝗶𝗰𝘀 — just 𝘴𝘰𝘶𝘯𝘥 like them. Our new phonetic attack shows that music and video generators like 𝚂𝚄𝙽𝙾, 𝚈𝚞𝙴, and 𝚅𝚎𝚘 𝟹 memorize songs through 𝙝𝙤𝙢𝙤𝙥𝙝𝙤𝙣𝙞𝙘 𝙨𝙝𝙖𝙙𝙤𝙬𝙨 — regurgitating

This new paper suggests AI music models are memorizing music they're trained on. Using phonetic substitutions in the lyrics, they get music that sounds strikingly similar to well-known songs. e.g. Lose Yourself, from Suno (with no genre specified by the user) 🧵 1/n

Morgan Stanley's Adam Jonas has released a video of how how AI and humanoid robots will change the world.

Turns out you don’t need the lyrics—just the sound of them. Our 𝐀𝐏𝐓 𝐚𝐭𝐭𝐚𝐜𝐤 shows AI music & video models regurgitate iconic songs through phonetic shadows.