AI and Llama3 with LightWeb3.0
Thanks to the open-source ollama from Meta, we managed to build a scientific database for the University of Louvain
What is Ollama 3?
Ollama is an open-source project that provides a powerful and user-friendly platform for running LLMs on a local machine. It bridges the complexities of LLM technology and the desire for an accessible and customizable AI experience.
What is WordPress?
WordPress (also known as WP or WordPress.org) is a web content management system. It was originally created as a tool for publishing blogs but has evolved to support publishing other web content, including more traditional websites, mailing lists and Internet forums, media galleries, membership sites, learning management systems, and online stores. Available as free and open-source software, WordPress is among the most popular content management systems—it was used by 43.1% of the top 10 million websites as of December 2023.
Artificial intelligence (AI) chatbots have opened up a world of possibilities for people to dabble in projects they might not have tackled due to lack of connectivity, but LightWeb solves this issue via NIZU.
While ollama3 in NIZU generated the PHP code for a WordPress plugin, it offered a few additional challenges.
Pierre AI helps the user easily find and interact with scientific articles on building and architecture, energy improvement, and renewables.
This frontend is built 100% in HTML and JS only and has no WordPress backend. However, the university content managers build the data in WordPress.
Generative AI holds immense promise for various development tasks, including creating a simple NIZU plugin.
However, not all AI models are created equal. Some stood out for their ability to create effective code with minimal prompting and troubleshoot code generated by other models.
The reason for using Ollama3 instead of ChatGPT is that it performs well in undergraduate-level benchmarks, scoring 82% on the MMLU 5-shot test, just behind GPT 4’s 86.4%. This suggests that while ChatGPT 4 leads in raw processing power, Llama 3 remains competitive in basic language tasks.
On more complex tasks requiring advanced reasoning, Llama 3 surprisingly edges out with a 35.7% score in graduate-level benchmarks, against GPT 4’s 39.5%. In coding-related evaluations, GPT 4’s superiority is evident again, scoring 85.9% in the HumanEval benchmark, surpassing Llama 3’s 81.7%. This reveals ChatGPT 4 as the preferable choice for programming and coding applications.