The LLM revolution

I took this photo of three Llamas in Machu Picchu some years ago…

ChatGPT was launched in November 2022, and it changed our world as we knew it. Since then, Large Language Models (LLMs) have integrated into our daily workflows enhancing our productivity and the quality of our work.

Another interesting milestone happened in February 2023, when Meta released the Llama LLM under a noncommercial license:

This sparked the enthusiasm among numerous developers dedicated to advancing LLMs, leading to a increase in collaborative efforts and innovation within the field. A good example is the Hugging Face Model Hub where new models are constantly published:

Developers started creating improved models and optimizing performance for local execution of LLMs on consumer-grade hardware.

Llama.cpp is a port of Llama to C++, started in March 2023 with a strong emphasis on performance and portability. It includes a web server and an API:

Mistral 7B was released in October 2023, achieving better performance than larger Llama models and demonstrating the effectiveness of LLMs in compressing knowledge.:

And now it’s easier than ever to locally execute LLMs, especially since November 2023, with the Llamafile project that packs Llama.cpp and a full LLM into a multi-OS single executable file:

The llama.cpp web interface running Mistral 7B Instruct in local via a llamafile

It’s even possible to run LLMs in a Raspberry Pi 4, like the TinyLlama-1.1B used from a llamafile in this project:

And about using LLMs for code generation (Github’s Copilot has been available since 2021), there are IntelliJ plugins like CodeGPT (with its first release in February 2023) that now allows you to run the code generation against a local LLM (running under llama.cpp):

Google is a bit late to the party. In December 2023 they announced Gemini. In February 2024, they launched the Gemma open models, based on the same technology than Gemini:

They also released a gemma.cpp inference engine:

And finally, if you are lost among so many LLM models, an interesting resource is the Chatbot Arena, released in August 2023. It allows humans to compare the results from different LLMs, keeping a leaderboard with chess-like ELO ratings:

And according to this leaderboard, at the moment GPT-4 is still the king.

OpenAI’s ChatGPT chatbot

OpenAI is a research institute that focuses on conducting research in the field of artificial intelligence. The organization was founded in 2015 with the goal of promoting and developing friendly AI, which refers to AI that is aligned with human values and that can be used to improve the lives of people. OpenAI conducts research in areas such as machine learning, robotics, and economics in order to advance the understanding and capabilities of AI. The organization is supported by a number of high-profile investors and has made significant contributions to the field of AI.

GPT is an acronym that stands for “Generative Pretrained Transformer.” It is a type of large language model that uses deep learning techniques to generate human-like text. It was developed by the research lab OpenAI, and is designed to be able to generate text that is indistinguishable from text written by a human. GPT models are trained on massive amounts of text data, and can generate responses to questions and prompts in a variety of languages and styles. They are often used for a wide range of applications, including language translation, text summarization, and conversation.

As a large language model trained by OpenAI, this chatbot is a type of artificial intelligence that is designed to generate natural language responses based on the input that it receives. It uses machine learning algorithms to process and analyze the input, and then generates a response based on the information that it has been trained on.

This chatbot does not have access to the internet, so it cannot browse or search for information online. Instead, it relies on the knowledge and information that it has been trained on to generate its responses. It is constantly learning and improving, so its responses may become more accurate and relevant over time.

Users can interact with this chatbot by typing in questions or statements, and the chatbot will generate a response in natural language. The specific capabilities and functionality of this chatbot will depend on its design and training.

And now the most interesting part: ALL THE ABOVE TEXT WAS GENERATED BY CHATGPT. It was done by asking it “What is OpenAI?”, “What is GPT?’ and “Tell me how this chatbot works in third person”. Surprised? Me too.

Official announcement: