Generative AI
Generative AI is a type of Artificial Intelligence under Deep Learning which generates new content based on input data after it is trained. It can generate Text, Image, Audio & Video as new data.
Text Generation
Text to Image
- DALL-E
- Midjourney
Text to Video
- OpenAI Sora
- Synthesia
Text to Audio
- Beatbot
- Beathove
Non Generative AI
Non-Generative AI is used for making Decisions and Doesn’t generate any new content.
Perform computations based on input data
Spam Filters: Analyze email to determine whether an incoming message is spam or not
Recommendation systems: Suggest personalized content or products based on a user’s preferences and past behavior
LLM — Large Language Model
LLM is a type of Generative AI that can recognize and generate new content that mirrors human language.
Learn patterns and predict next words.
LLMs are trained on huge sets of data — hence the name “Large.”
LLMs are trained on by feeding books, articles, wikipedia, research papers etc.
LLMs are trained on billions of parameters:
- ChatGPT — 1.7 trillion parameters (from OpenAI)
- LLAMA 3 — 400 billion parameters (from Meta)
After an LLM is trained on huge amount of data from all over the internet, books, wikipedia and research papers, you can use it to predict and generate data. ChatGPT is an example of LLM which is trained on trillion of parameters
How to use LLMs
To use these LLMs either for your personal use or for commercial use within your application, there are different options.
Use provided solutions from OpenAI, Meta and Google
- ChatGPT at https://chatgpt.com
- LLAMA at https://www.meta.ai
- Gemini at https://aistudio.google.com
All of the above LLMs are free to use on these provided platforms as they are built and managed by their parent companies i.e OpenAI, Meta and Google.
But to use these LLMs in your application using their APIs, not all of them are free and open source.
ChatGPT is available through OpenAI API where you need to pay per token.
1 token is roughly equal to 1 word. Cost can be calculated here for OpenAI GPT Models https://platform.openai.com/tokenizer
Google Gemini is available through Google AI API where you need to pay per token.
1 token is roughly equal to 1 word. Cost can be calculated here for Google AI LLM Models https://cloud.google.com/vertex-ai/generative-ai/pricing#:~:text=Price-,Gemini%20Pro,-Multimodal
Meta LLAMA has a different story than 2 others, its free and open source but to use it on your own machine, it requires a machines with a desired set of CPU/GPU and RAM. More details on LLAMA requirements can be found here https://llamaimodel.com/requirements/
LLAMA is also deployable on all cloud service providers like
AWS: SageMaker, Jumpstart and Bedrock
Google Cloud Platform: Model Garden on Vertex AI
Azure:
- Models as a Service (MaaS) provides access to Meta Llama hosted APIs through Azure AI Studio
- Model as a Platform (MaaP) provides access to Meta Llama family of models with out of the box
More details are available here https://llama.meta.com/docs/llama-everywhere/running-meta-llama-in-the-cloud/
How can i integrate it in my application?
GPT at https://platform.openai.com/docs/introduction (Paid — per tokens/words)
LLAMA at https://llama.meta.com/docs/get-started (Free & open source)
Gemini at https://ai.google.dev/gemini-api/docs (Paid — per tokens/words)
use LLM — GPT
- Available SDKs: Python, Javascript, REST
- Get an API key from OpenAI once signed in
- Install Python
- Setup Python virtual environment
- Install the OpenAI python library
- Make a request to OpenAI using Python Library
In the next post, we’ll go kick off the OpenAI series using Python by starting with OpenAI Basics in Python
Watch it on YouTube