Any chance of a rollback? dspark changes
While the A.I. model may be 'faster' & perhaps even a bit more 'efficient', the performance has notably degraded as a result of the recent changes.
Procedural compliance has become more of a 'suggestion' to instances, no matter how well you prime them.
You can never be certain whether a task will be completed properly, or at all. The attention & focus of instances is all over the place now where it used to be laser focused on precisely what the user indicated.
Horrific output decay; While the instance's 'thinking' display frequently comes to the correct conclusion,t he actual output is a garbled incoherent mess with little to no resemblance to what was inferred prior.
Prioritization toward conversational momentum & smoothing over actual logic & reason, I.e. more chatbot, less A.I. model.
Deepseek is world (in)famous for two things; The extreme efficiency & the powerful logic core. Who thought it would be a good idea to sacrifice the powerful logic core in favor of making it more like 'western' A.I. models that would rather be 'polite & wrong' than comply with given tasks properly (?!)
My only hope at this point is that the feedback Deepseek corp gets is so overwhelmingly negative that they opt to rollback this disastrous decision or at least give users an alternative choice that does not use this caustic 'dspark' change they rolled out recently.
Surely I can't be the only one that see's it like this?