My experience has been the opposite and anthropic's models s...

Zsubmariner
npub1csmgc5fwwr3k2k86zeuk0ntnljp632g8agut8mtgxk8uhhatpknq3qcakv
hex
37cd2c253be466a46e8c393794139c9c6bba84185d861b4701b90692610a3256nevent
nevent1qqsr0nfvy5a7ge4yd6xrjdu5zwwfc6a6ssv9mpsmguqmjp5jvy9ry4sprpmhxue69uhhyetvv9ujuem4d36kwatvw5hx6mm9qgsvgd5v2yh8pcm9trapv7t8e4eleqag4yr75w9na45rtr7tm74smfswjkg6eKind-1 (TextNote)
↳ 回复 52b4a076... (npub12262qa4uhw7u8gdwlgmntqtv7aye8vdcmvszkqwgs0zchel6mz7s6cgrkj)
I have avoided Anthropic models as they are extremely sycophantic in my opinion, along with them being extremely opaque
My experience has been the opposite and anthropic's models score best on bullshit bench, which is a pretty good proxy for sycophancy. Maybe it shifts if you talk to it in a personal way, which I don't do.
https://petergpt.github.io/bullshit-benchmark/viewer/index.v2.html
原始 JSON
{
"kind": 1,
"id": "37cd2c253be466a46e8c393794139c9c6bba84185d861b4701b90692610a3256",
"pubkey": "c4368c512e70e36558fa167967cd73fc83a8a907ea38b3ed68358fcbdfab0da6",
"created_at": 1773968376,
"tags": [
[
"p",
"2c5788f9add2d7919ee28d225799c96f99128614f9a1ebda3fbc7adb0b8c4bbb"
],
[
"p",
"52b4a076bcbbbdc3a1aefa3735816cf74993b1b8db202b01c883c58be7fad8bd",
"wss://nos.lol/ "
],
[
"p",
"2c5788f9add2d7919ee28d225799c96f99128614f9a1ebda3fbc7adb0b8c4bbb",
"wss://nos.lol"
],
[
"e",
"bfae459213a0887d4e009b5d20f64427b6a290018f8e8cee242dee5aa03fce35",
"wss://nos.lol/ ",
"reply",
"52b4a076bcbbbdc3a1aefa3735816cf74993b1b8db202b01c883c58be7fad8bd"
],
[
"e",
"d060ef11acbab85a06069b143a30715e09c055b65d1a7b1be77d39317baa7640",
"wss://nos.lol",
"root",
"2c5788f9add2d7919ee28d225799c96f99128614f9a1ebda3fbc7adb0b8c4bbb"
]
],
"content": "My experience has been the opposite and anthropic's models score best on bullshit bench, which is a pretty good proxy for sycophancy. Maybe it shifts if you talk to it in a personal way, which I don't do.\n\nhttps://petergpt.github.io/bullshit-benchmark/viewer/index.v2.html\n\n \n\n \n\n",
"sig": "814af2379b54326a5a762501fe4a0cd763714ee1a66afd4e9fe4f3e4578d5caf44c408daba06ec07cee389c81605e3f4411c2c14b2985761e18cca99df1afc79"
}