Ask HN: Anyone spent big money finetuning OpenAI models that are now deprecated?
Is this just the cost of doing business? I'm afraid to train new models for fear of this happening again.
My take is the following: - To fine-tune, we can only use 3.5. - And it is not changing a lot as the OpenAI focus on 4+ models.
So, there is no point in fine-tuning once again as new 3.5 model released. We only re-fine-tune when we have new data and approaches; then we can fine-tune on new versions of the model.