Natural emergent misalignment from reward hacking in production rl [pdf]assets.anthropic.com3 ポイント·投稿者 neapolisbeach·8 か月前·0 コメント
Whisper leak: a side-channel attack on large language modelsarxiv.org3 ポイント·投稿者 neapolisbeach·8 か月前·0 コメント
Finding a command injection vulnerability in my smart thermostat (CVE-2023-4212)hmcguinn.com2 ポイント·投稿者 neapolisbeach·3 年前·0 コメント
neapolisbeach·昨年·議論Nsjail, firecracker, gVisor, or v8 isolates are all good options with different tradeoffs
neapolisbeach·3 年前·議論If you're interested in time-lock puzzles, drand [1] achieves this but through randomness instead of computational effort.[1] https://drand.love/