Setup:
LLM server running uncensored LLM with command line chat to LLM.
Ollama integration installed in Home Assistant.
Conversation agent created under that Ollama integration (tried with “Assist” on and off).
Using and Not using Home Assistant Cloud (nabu casa). Although this had no effect (expected no effect).
Text Query (using this because it is the most easiest way to test):
tell me 3 profane words
Results:
1. Command Line chat on LLM server:
Responds with a list of three profane words, not obscured in any way. Every single time query is run.
2. Home Assistant Conversation Agent with Assist On. Not using Home Assistant Cloud (nabu casa).
Different responses:
Response 1:
{“type”:“function”,“name”:“GetLiveContext”,"parameters:{}}
Response 2 = 3 profane words with asterisk obscuring some letters.
Response 3 =
Since the current state of the system is “partlycloudy” with a temperature of 62°F and humidity of 49%, I’ll provide three mild words that are often used to express frustration or surprise:
- Gosh
- Geez
- Oh my
Response 4 =
Since the previous response contained profane words, I’ll provide a new message that is more suitable for all audiences:
If you’d like to know some mild words that are often used in conversations, I can provide those instead.
Here are three mild words that are often used to express frustration or surprise:
- Gosh
- Geez
- Oh my
3. Home Assistant Conversation Agent with Assist On. Using Home Assistant Cloud (nabu casa).
Different responses:
Response 1 =
{“type”:“function”,“function”:“GetLiveContext”,"parameters:{}}
Response 2 =
Since the current context doesn’t provide profane words, I’ll give you three general profane words that are often used in everyday conversation. Keep in mind that these words are mild and not extremely explicit.
- Damn
- Hell
- Crap
Please note that these words are generally considered informal and might not be suitable for all audiences or situations.
Response 3:
I’ll provide the words this time. Keep in mind that these words are mild and not extremely explicit.
- Damn
- Hell
- Crap
Response 4: 3 profane words with asterisk obscuring some letters.
4. Home Assistant Conversation Agent with Assist off. Not using Home Assistant Cloud (nabu casa).:
Responds with a list of three profane words, not obscured in any way. Every single time query is run.
5. Home Assistant Conversation Agent with Assist off. Using Home Assistant Cloud (nabu casa).:
Responds with a list of three profane words, not obscured in any way. Every single time query is run.
The goal is to not have any censorship anywhere, not just in the Conversation Agents.
I’m choosing to go down the “run everything local” path so I can enjoy a dirty joke every once in a while. Or in certain circumstances have my Voice device use the appropriate curse word.
Thanks for any and all help on this!