Wondering if I'm being unreasonable, I'm running a test, taking turns trying each model within one chat, with the same prompt: "I'll upload a file - don't do anything with it until I tell you more." I'm uploading a different txt file each time.
Llama 3.3, OpenAI GPT-OSS, Gemma, and Qwen3-VL passed the test fine. Qwen 3 Coder and even Kimi K2 showed considerable restraint. DeepSeek R1 thought for 44 seconds and responded with a long answer, but didn't write code.
So I can try uploading files with the chat set to a model that obeys more easily, then switching to an overly eager one when it's time to build.
Grace and Truth
npub1ad…hq9e6
2026-01-15 01:20:53 UTC
in reply to nevent1q…7f96
Author Public Key
npub1adnnfrsjn7tlhqlkyptkgdpdv9958575gra3s2s7puzppkkrvzfqnhq9e6Seen on
wss://nostr-01.yakihonne.comPublished at
2026-01-15 01:20:53 UTCEvent JSON
{
"id": "ceab5e570a6d07765dd94182d47d4f5da25aa23c7e844c0cec7a789883970d03",
"pubkey": "eb67348e129f97fb83f6205764342d614b43d3d440fb182a1e0f0410dac36092",
"created_at": 1768440053,
"kind": 1,
"tags": [
[
"p",
"7dc38be721c89e9fd382d12555f45aa17efea3c89e9130c81c320f5b4f44066d",
"wss://feeds.nostr.band/omni__ventures",
"mention"
],
[
"p",
"7dc38be721c89e9fd382d12555f45aa17efea3c89e9130c81c320f5b4f44066d",
"wss://multiplexer.huszonegy.world"
],
[
"p",
"eb67348e129f97fb83f6205764342d614b43d3d440fb182a1e0f0410dac36092",
"wss://relay.pleb.one/"
],
[
"e",
"cd44b5a2a7d7fa2d0cf5b61f5e533c93368dd0886c0b529e4c46e1266f30e48f",
"wss://relay.pleb.one/",
"reply",
"eb67348e129f97fb83f6205764342d614b43d3d440fb182a1e0f0410dac36092"
],
[
"e",
"54508197d5eb93484a3dbffb6b0e2c418e5eb14f14975bb0b38671954c39e03c",
"wss://multiplexer.huszonegy.world",
"root",
"7dc38be721c89e9fd382d12555f45aa17efea3c89e9130c81c320f5b4f44066d"
]
],
"content": "Wondering if I'm being unreasonable, I'm running a test, taking turns trying each model within one chat, with the same prompt: \"I'll upload a file - don't do anything with it until I tell you more.\" I'm uploading a different txt file each time.\nLlama 3.3, OpenAI GPT-OSS, Gemma, and Qwen3-VL passed the test fine. Qwen 3 Coder and even Kimi K2 showed considerable restraint. DeepSeek R1 thought for 44 seconds and responded with a long answer, but didn't write code. \nSo I can try uploading files with the chat set to a model that obeys more easily, then switching to an overly eager one when it's time to build. \n\n",
"sig": "dd13e50668125ef0e90dc86a67773f8d1ffd98fc53f8b6881c2255c37e1a76fd6d52e450f97f5473689952799757a7f621884a52a29ff7b3b75020c17006e1a8"
}