etalab-ia / albert

Backend ressources for Albert. Albert is a conversational agent that uses official French data sources to answer administrative agents questions.
MIT License
109 stars 7 forks source link

Conflicting with OpenAI's terms of use? #6

Open Alkarex opened 2 months ago

Alkarex commented 2 months ago

Hello, I am wondering whether Albert's models such as guillaumetell-7b are conflicting with OpenAI's terms of use:

https://openai.com/policies/terms-of-use

What You Cannot Do [...] Use Output to develop models that compete with OpenAI.

Indeed, guillaumetell-7b is based on OpenHermes-2.5-Mistral-7B, which is trained on OpenHermes-2.5, which in turn is based on plenty of OpenAI GPT-4 outputs such as GPT4-LLM-Cleaned (with a problematic license). And the resulting model clearly competes with OpenAI own services.

So it looks like there is a risk of domino effect there, where everything can fall when OpenAI starts enforcing their own terms of use.

Any thoughts?