Is there any hope that something like unsloth studio will let people retrain models continually so that, when the day comes that there are no good open models being released, something like the final generation of qwen whatever can continue being relevant into the future?
It would likely fall on the harness (and possibly local datasets) to encourage that the model reach out for up to date information instead of relying on it's own internal "knowledge".
We already see a lot of that with `web_search` tooling, so I imagine it would just become more essential to have tools like that.