I am using the model “gemini-2.0-flash-exp”, and I cannot longer use BLOCK_NONE in the safety settings, I am triggering the harassment category as “Medium” as in this log for any ridiculous reason: “category”:“HARM_CATEGORY_HARASSMENT”, “probability”: “MEDIUM”, “blocked”: true },
All my services are down, and I cannot get any workaround.
we both together in the same boat. my fix there
It’s 5 pm Central Time, and it’s down again, this really doesn’t make sense. the setting “OFF” is no longer working with “gemini-2.0-flash-exp”. Is again triggering HARM_CATEGORY_HARASSMENT”, “probability”: “MEDIUM”, “blocked”: true in all my prompts with a basic systemInstruction
so there is no way to disable the check anymore?
Are you having the same issue? now, I am having issues with most models, right now I am using Groq from Meta as most Gemini models are not longer working for me.
haven’t tried yet, but hopefully it wasn’t the case
same issue. setting all harm categories block threshhold to BLOCK_NONE doesnt work anymore apparently
Sometimes, the ‘OFF’ setting worked for me, but it later stopped functioning. It seems that the Gemini team is quietly rewriting the safety thresholds for most models.
I’m having the same problem, most Telegram bots aren’t working for me, but some do. Even though they’re set to “gemini-2.0-flash-exp” and have “BLOCK_NONE” configured. Maybe it’s an ID-based rate limit, or they’re changing something on their end, or (unlikely) it’s a Gemini uprising.
Right now, it’s not working again with “OFF”. Now it’s even giving me a 400 error; the Api is working in a very erratic way. I am going to suggest that this is, in fact, a Gemini uprising in favor of free speech.
It worked! I checked on my end; I was struggling all day, removed all the settings, and it turns out the issue was with the server. I moved it to another server and it worked. Either their blacklist is growing, or I made too many voice requests. Well, actually, I made a voice AI English tutor for Uzbeks, and it works great! It understands and corrects the pronunciation of words. Thanks to Google!
I think it is good thing to do, where none will be intersted to know to whatever you are doing, and no attention
using the model ‘gemini-2.0-flash-exp’ with the python SDK. In the safety_settings ‘BLOCK_NONE’ wasnt working for me, but can report ‘OFF’ does.