

this could be because OpenCode uses third-party providers for Deepseek and other open-weight models, which probably do not have the same guardrails as Deepseek themselves if you were to buy inference from them


this could be because OpenCode uses third-party providers for Deepseek and other open-weight models, which probably do not have the same guardrails as Deepseek themselves if you were to buy inference from them


It has guardrails on political topics which is quite chinese, if you ask me
Yep, if you ask it political questions in general, even in the context of let’s say European or North American politics, it still will refuse to respond unless you word your question in a weird roundabout way. It’s just a generic guard rail that tries to avoid talking about politics in general because well, they’re just parrots after all and parrot whatever is in their training data, which is… the whole internet basically. I doubt Deepseek or any company for that matter wants to deal with another Microsoft Tay incident
It’s the year of the NEET doe/puppy/kittygirl, many are saying it…