How do I access the Zephyr-Mistral LLM model for uncensored chat on You.com?

You.com typically employs proprietary commercial Large Language Models (LLMs) to power the AI on our site. These models are known for producing high-quality text and driving sophisticated features, with advanced moderation tools integrated to ensure output remains appropriate and Safe For Work (SFW).

However, when users set their Safe Search to ‘Off’, these moderation features might be overly restrictive, leading to instances where You.com may not summarize search results or may experience a reduction in functionality.

For users who opt for the “Safe Search: Off (uncensored chat)” setting, You.com switches to a less-censored, open-source LLM. Currently, we are utilizing the Mistral 7B model, specifically the Zephyr 7B alpha model, which is an instruct fine-tune of Mistral. We aim to keep this choice updated with the best open-source models available. This LLM adheres to what can be described as “community standards” across the web, meaning it may still refuse some queries, but less frequently than commercial models. Using an open-source model also allows You.com to apply our moderation to modify its default behavior if needed.

It’s important to note that if you request Not Safe For Work (NSFW) content from this LLM, it will likely provide such content, similar to other search engines with Safe Search turned off.

While the open-source model is less censored than our regular commercial AI, it is also less powerful and may produce lower-quality text with more errors. We recommend not using the Safe Search Off setting as your default. Instead, switch to it only when necessary, like when encountering refusals from our regular AI.

To remind you of this setting, we’ve added a notification at the end of its outputs, indicating that you are using the less-censored LLM option.