Degraded Performance on gpt-4.1-mini ft

@vb
You can see some here are 2000t (which was the limit I applied). Looks like almost all the errors have to do with cases where there are no angel investors… The training set is imbalanced towards this as well (more no angel investors than non). Again its been fine for months and just started acting up.

Json Schema
chatcmpl-Cld6a6FUjJ4LdWGuEKjLCoRHSKA7z

chatcmpl-ClPyPLpp4Cp9DopvDj3ElUMhFxZ8d

chatcmpl-ClPyNlUI5O8QGwOSXA3zYKvaQJhWj

chatcmpl-Cld66lS1t4m2qdPlJBTh603qk8Zcy

chatcmpl-CldsWog5gpqkR6pmZaXC9bxNxYOap

Text

chatcmpl-ClPzmq2GwofLefXXIIvq2oPfCmWXR

2 Likes

This one here just broke my data pipeline - chatcmpl-ClmCMYTLJp9MqF9wwQqu7ss2pwosV
Outputted this…
{
“investors”: 847
}

1 Like