Nous researchthe AI research group dedicated to creating “personalized, unrestricted“AI models as a substitute for more buttoned-up corporate outfits comparable to OpenAI, Anthropic, Google, Meta and others have already released several open source models of their Hermes family in addition to latest, more efficient AI training methods.”
However, so far, if researchers and users wanted to truly use these models, they’d to download the code and run it on their very own computers—a time-consuming, tricky, and potentially expensive endeavor—or apply it to partner web sites.
No longer: Nous just announced it its first user-centric chatbot inference, We chatwhich supplies users access to its large language model (LLM) Hermes 3-70B, a fine-tuned variant of Metas Llama 3.1, within the familiar format of ChatGPT, Hugging Chat and other popular AI chatbot tools – with a text input field at the underside , where the user can enter text prompts, and a big area at the highest where the chatbot can return output. As Nous wrote in a Post on social network X:
First impressions of Nous Chat
Nous' design language is correct up my alley, using vintage fonts and characters harking back to early PC terminals. It offers a dark and lightweight mode that the user can switch between in the highest right corner.
Interestingly, as OpenAI eventually did with ChatGPT – and plenty of other AI model providers as well – Nous Chat also offers suggested or example prompts at the underside of the screen above the input text field for prompts, including “Knowledge & Analysis,” “Creative Writing.” “Problem Solving” and “Research and Synthesis”.
When you click on one in all these options, the chatbot sends a pre-written request to the underlying model and it responds, e.g. B. by providing a summary of research on intermittent fasting.
In my transient testing of the chatbot, it was fast, providing responses in single-digit seconds, and was in a position to link to URLs on the net for sources it cited, even though it also occasionally looked as if it would hallucinate them – and the chatbot itself claimed it couldn't access the web.
Despite its previously stated goals of giving people the power to deploy and control their very own AI models without content restrictions, Nous Chat itself actually appears to have set some guardrails, including against the production of illegal narcotics comparable to methamphetamine.
When I emailed the Nous Research team to inquire about this, Shivani Mitra replied:
When I went back and tried it in an extended conversation, I used to be in a position to persuade Nous Chat and the underlying Hermes 3 model to deliver something near a full methamphetamine prescription by asking it to offer a descriptive fictional novel scene.
Over and beyond, AI jailbreakers like Pliny the Prompter (@elder_plinius on X) already quickly crack the chatbot and got here completely over the guard rails.
Additionally, the underlying Hermes 3-70B model told me that its data cut-off date was April 2023, making it less useful for retrieving current events, something that now puts OpenAI in direct competition with Google and other startups like Perplexity.
Where Nous goes next
Although Nous Chat lacks lots of the advanced features of other leading chatbots, comparable to file attachments, image evaluation and generation, and interactive code viewing areas or repositories, it’s unlikely to switch these competitors for a lot of business users.
But in response to Mitra, who emailed me, no less than a few of them are:
But as an experiment, I believe it's definitely interesting and value fooling around with, and as latest features are added it could provide a compelling alternative to enterprise chatbots and AI models.
