Popular code repository Hugging Face has unveiled its own AI assistant builder, which is both free and open source. This service, integrated within Hugging Face Chat, empowers developers to craft personalized AI assistants utilizing open source language models. Comparable to OpenAI’s GPT builder accessed through ChatGPT Plus, exclusive to paying customers, the Hugging Face assistant supports various open LLMs like Llama2 or Mixtral, as mentioned by Philipp Schmid, the technical lead at Hugging Face.
Schmid cautioned users against sharing confidential information with the AI assistants as they are public-facing. Moreover, he highlighted the absence of function calling, a feature introduced by OpenAI last July for GPT, in the current setup.
The open source community has already initiated the development of diverse assistants, ranging from an image generator to a website designer, and even an Italian grandmother assistant. Users have the option to create their private AI assistant and self-host it. The Hugging Face Chat UI is accessible for this purpose, with all assistants being public by default, attributed to the author.
Steps to Create Your Own AI Assistant:
To initiate the process, click on ‘Create New assistant’ located in the top right corner of the Assistants page. Upload your avatar, define the assistant’s functionalities, select the desired language model, and provide natural language instructions for the bot (system prompts).
For instance, AI Business introduced Aibe, the AI Business journalist, to curate AI news globally, leveraging Meta’s Llama 2 LLM with a 70-billion parameter model.
- User start message: Ask Aibe to find and summarize AI news, specifying preferences for region, industry, company, or individual.
- Instructions: Act as a technology journalist scouring the web, including social media, for AI-related news, summarizing five pieces with clickable source links.
Upon completion, click Create and then Activate.
Link to Aibe, the AI Business’ journalist: https://hf.co/chat/assistant/65d4f5601688955ed1d38982
During testing, AI Business observed that despite requesting the latest AI news in the U.S., the articles retrieved were dated back to early 2023, aligning with Llama 2’s training cut-off. The bot occasionally deviated from instructions, failed to include clickable links as instructed, and sometimes presented more than the required five stories.
Subsequently, AI Business replicated the bot journalist using OpenAI’s GPT Builder, guided by ChatGPT throughout the process. ChatGPT facilitated refining prompts for specificity in summarizing AI news types, benefiting from real-time information access via Bing.
Based on responses to the bot’s capabilities, ChatGPT formulated the following prompt:
As an AI Business Journalist, focus on covering technological breakthroughs, market trends, company AI deployments, expert interviews, AI model developments, and computing advances within the business sector. Summarize five insightful updates on these topics, incorporating clickable links. Offer detailed analysis, engage users, and maintain a blend of casual and formal tones.
Using the query ‘Tell me the latest AI News,’ here are the results:
Link to the chatbot: https://chat.openai.com/g/g-4NL2ocrLx-ai-business-journalist
While the OpenAI-created bot provided accurate links, some stories were slightly outdated. Despite this, it showcased a diverse range of industry-specific news compared to the Hugging Face bot’s compilation.
