claudePOST
Claude bans are hard to test against, since new accounts add too many variables
The author argues Claude bans are hard to reproduce because each new account brings many variables at high cost, while OpenAI rarely bans but quietly routes users to a smaller model, which allows controlled tests.
LotusDecoder
@LotusDecoderArchived 16 days agoOriginally posted 2026.10.16
LotusDecoder @LotusDecoderOriginally posted 2026.10.16
claude 封号真是一个经久不衰的话题, 大概从测试角度,难以做回归。 封了就没了,再换一个新号,又有大量变量。 对于个体用户来说,拥有大量claude特征相同的新号, 成本是过高的。 openai 就好得多了,很少封号,但还是有。 风控措施是悄咪咪暗改模型, 路由到一个小模型里。 所以根据回答速度和质量, 可以改变运行环境的变量, 做对照实验,测出是哪个因素占大头。
x.com/lotusdecoder/status/2110989951764660704Archived 2026.10.17 10:12