Major AI Platforms and Tools

by William Martin

These are the platforms we'll be discussing in this course. Each are chosen for a few reasons:

Significance – How influential is the platform or toolkit (and its parent company, if any) in the AI space? Foundational
Variety of capabilities – These platforms can each perform a large variety of tasks and support training, fine-tuning, inference, etc.
Variety of tooling – These platforms have different ways to access its capabilities, like playgrounds, an SDK, etc.
Likelihood of longevity – Robust platforms and tools that are likely to be around for a long while, so the skills you build will last.

The ones below aren't the only ones worth using or trying, but there are so many that it's not really possible to cover even a significant number of them. The ones here provide enough of a foundation to be capable and effective with AI.

Hosted Platforms

Hosted Platform: OpenAI

It's impossible to talk about AI these days without mentioning OpenAI. This company and its offerings have popularized the newest generative AI capabilities, and their platform is still the state-of-the-art and the standard to access large language models and other capabilities.

OpenAI have two product lines: ChatGPT and the OpenAI API (application programming interface).

ChatGPT

ChatGPT is a chat application that utilizes OpenAI's underlying models, GPT (which stands for Generative Pre-trained Transformer). "Transformer" refers to an architecture we'll discuss in the next sequence.

API

The OpenAI API is a way to utilize the capabilities of OpenAI in a more open and flexible way. More than just the GPT models, we can access vision capabilities, text-to-speech, speech-to-text (i.e. transcription), embeddings, and more. We'll be using the API to understand how to leverage some of the most advanced AI abilities, to enable machines to "speak human," and to enable LLMs to control systems outside of the AI itself.

We'll use both products in this course, with an emphasis on the API. For ChatGPT and this course, the free account is sufficient.

GPT vs ChatGPT

It's important to distinguish between GPT and ChatGPT.

ChatGPT is an app, available on the web and through a mobile app, that provides a chat-like user interface, features like voice conversations, memory, and storage of chats. With these tutorials, the public version of ChatGPT, found at https://chatgpt.com/ will be sufficient.

GPT, however, is a class of models for text completion (and more). OpenAI provide several versions of GPT, each with different performance characteristics and capabilities, like structured output and function calling. We'll talk more about them later, but the key point is that we can access GPT models multiple ways, and one of them is ChatGPT, the app.

Task: Set up billing for the API

The API requires a billing account. The cost is usage-based, but we won't be sending many requests in these tutorials. If you decide to experiment more by writing code against the API, be careful of your usage!

Task: Set up a new Project for the Smorgasbord tutorials.

OpenAI's platform product can be confusing. Here are the concepts to keep in mind:

Organization – An "organization" is an account-level construct intended to represent companies or teams. By default, you will have a single organization. I recommend naming this one "Personal."

Project – An organization can have multiple "projects," which are constructs that represent a bundle of work, like an app. There is always a single project called the "default project" that you cannot delete. By creating additional projects, you can add to them API keys and billing limits.

Add a new project and set a low billing limit (e.g. $10 per month) like this:

Hosted Platform: Google Vertex AI Studio

Google Vertex AI is a product line inside Google Cloud, one of the three major public cloud providers (alongside Microsoft's Azure and Amazon's Amazon Web Services, AWS). It provides both a web-based dashboard and APIs and SDKs (software development kits) that we can use for Google's AI models, in particular, Gemini. Google also provides a marketplace where third-party providers give access to their AI models and services inside Google Cloud.

The advantage of Google Vertex AI is that we can test many different modalities in the same web UI. We can easily ask questions of videos alongside summarizing documents, or at least, that's the promise.

Hybrid Platform: Hugging Face

Hugging Face is a different kind of AI company from the ones above. You can think of them as a kind of "GitHub for AI." Many companies, including major tech companies like Meta, and individual teams and researchers publish their models here.

In addition to a public repository for models, Hugging Face provides multiple ways to test and deploy them models, including web-based chat, SDKs, rentable cloud environments (that they call "Spaces"), and more.

In particular, these SDKs (accessible with Python and most with JavaScript) are quite useful since they provide a single consistent way to use nearly any AI model on their platform.

They have a number of great learning resources and cover a wide range of tasks.

ML Frameworks in Code

For the code-savvy, the following frameworks are the state-of-the-art with regards to using machine learning to implement AI methods.

Tensorflow

Tensorflow is a popular collection of machine learning libraries and frameworks from Google. It’s intended for anyone from beginners to experts, for multiple platforms including web (Tensorflow.js) and mobile, and for scalable production deployments.

We'll look at Tensorflow later. For now, find online tutorials here: https://www.tensorflow.org/tutorials

PyTorch

PyTorch is a Python library from The Linux Foundation similar to Tensorflow, described as an "optimized tensor library for deep learning using GPUs and CPUs."

PyTorch is supported on multiple operating systems and platforms, but is exclusively a Python library.

We'll also take a short look at PyTorch later, but you can find the official tutorials here: https://pytorch.org/tutorials/