Quickies: Part 1 — AI/LLM
We all happened to see this huge wave of conversation around ChatGPT. When I heard about it, it was a chatbot that can have a human dialog in an intelligent way. Like everyone else, I made my share of poking ChatGPT with questions and doubts. It was good. I had my own fear but as a technologist, I see the opportunity to change the future with this tech. Let’s have an informal catchup for anyone wanted to have a head start with this article.
AI — Artificial Intelligence — A field where computers can perform tasks that normally require human intelligence. Examples are Speech recognition, decision-making, language translation, etc
Humans are sophisticated and complex with any of those factors mentioned above. But AI has been around for a long period and they were trying to understand and mimic human intelligence. We heard about Machine Learning for a while where we can have a large set of data that can be fed to computers and they can help us predict certain things with complex math. But ChatGPT started generating the next words or sentences based on human input instead of the computer language or code for other AI advancements. This is huge. It’s this moment where your kid started talking the first few words and you are excited about it! We are in the same stage. ChatGPT and others will start making many improvements which will create human-like behaviors in several fields in the near future. This brings up this AI development called “Generative AI”. It generates text or images or video at this moment.
Generative AI takes human input as a “Prompt”, processes it through a “Model” and generates “Completion”. ChatGPT uses a model called GPT-4 (as of now) which allows chat mode to interact and get completions in the form of answers, or sentences, or even computer code. There are several other models or “Large Language Models” available: GPT, BERT, FLAN-T5, LLaMa, BLOOM, PaLM, etc.
Each of these LLMs is trained to do certain tasks.
- Essay Writing
- Summarization — Takes a large text and can summarize with few sentences
- Translation — Between languages or computer code
- Information Retrieval — Extract names, objects, etc. from a paragraph
- External Connectivity — Extend a model capability with new data by training with samples
These LLMs have been trained with millions to trillions of parameters where they are focused on the above tasks. The more they have it, they understand our conversation context better. But to work on a specific area, we can have models trained only in millions of parameters (like a kid who passed his/her first grade and also plays piano at the age of 5).
We are at the beginning of evolution and it will be interesting how they make our life better the way other revolutions did.
To learn more about LLM/Transformer, click here: Part 2
Disclaimer: This is not generated by an AI bot. Also, a lot of these were learned through the DeepLearning.ai course at Coursera. It’s a great course. The title image was generated by DALL-E through the Bing chatbot.
