Chatbots identify themselves very often in casual/non-technical chats AFAIK -- for example, when people ask it for its opinion on something, or about its past.
Re:sed, I'm under the impression that most chatbots are pretty pure-ML these days. There are definitely some hardcoded guardrails, but the huge flood of negative press early in ChatGPT's life about random weird mistakes can be pretty scary. Like, what if someone asks the model to list all the available models? Even in this replacement context, wouldn't it describe itself as "GLM Opus"? Etc etc etc.
It's like security (where absolute success is impossible) but you're allowed to just skip it instead of trying to pile Swiss cheese over all the problems! You can just hookup a validation model or two instead and tell them to keep things safe and enforce XYZ, and it'll do roughly as well with way less dev time needed.
After all, what's the risk in this case? OpenAI pretty credibly accused DeepSeek of training R1 by distilling O1[1], but it's my understanding that was more for marketplace PR ("they're only good because they copied us!") than any actual legal reason. Short of direct diplomatic involvement of the US government, top AI firms in China are understandably kinda immune.
Re:sed, I'm under the impression that most chatbots are pretty pure-ML these days. There are definitely some hardcoded guardrails, but the huge flood of negative press early in ChatGPT's life about random weird mistakes can be pretty scary. Like, what if someone asks the model to list all the available models? Even in this replacement context, wouldn't it describe itself as "GLM Opus"? Etc etc etc.
It's like security (where absolute success is impossible) but you're allowed to just skip it instead of trying to pile Swiss cheese over all the problems! You can just hookup a validation model or two instead and tell them to keep things safe and enforce XYZ, and it'll do roughly as well with way less dev time needed.
After all, what's the risk in this case? OpenAI pretty credibly accused DeepSeek of training R1 by distilling O1[1], but it's my understanding that was more for marketplace PR ("they're only good because they copied us!") than any actual legal reason. Short of direct diplomatic involvement of the US government, top AI firms in China are understandably kinda immune.
[1] https://www.bgr.com/tech/openai-says-it-has-evidence-deepsee...