É isso. A mesma opinião que compartilho aqui sobre os modelos de IA: O quanto os modelos que precisam implementar guardrails por pressão governamental etc tendem a ficar pra trás do que os pesquisadores de segurança precisamIt’s becoming obvious that relying on frontier models is getting to be risky business. Either the model gets export controlled, or it gets nerfed with hidden guardrails. The second one is the more annoying problem for defensive cybersecurity work. When a model like Fable outright blocks anything security-related, that’s at least transparent, because we know upfront it doesn’t work for cybersecurity. The real problem is hidden guardrails, a frontier model quietly stops working for a use case and we can’t tell why, and sometimes an older version performs better than the new one (more on that later in the blog).

É isso. A mesma opinião que compartilho aqui sobre os modelos de IA: O quanto os modelos que precisam implementar guardrails por pressão governamental etc tendem a ficar pra trás do que os pesquisadores de segurança precisam

It’s becoming obvious that relying on frontier models is getting to be risky business. Either the model gets export controlled, or it gets nerfed with hidden guardrails. The second one is the more annoying problem for defensive cybersecurity work. When a model like Fable outright blocks anything security-related, that’s at least transparent, because we know upfront it doesn’t work for cybersecurity. The real problem is hidden guardrails, a frontier model quietly stops working for a use case and we can’t tell why, and sometimes an older version performs better than the new one (more on that later in the blog).

https://www.hacktron.ai/blog/open-weight-models-vs-frontier-models