A Quick Summary of the major LLM AI Players and Systems

I had a phone call with March L. who was trying to work out a good way to get a foothold in understanding the Generative AI and LLM space as it has unfolded in the past year or so. I spoke off-the-cuff for while but then told her I'd write up a quick braindump for her and share it in on LinkedIn, since dozens people would jump on the comment thread to correct and clarify anything I got wrong. I have listed the players in what I consider the order of industry influence.

The OpenAI Camp

OpenAI launched the first salvo of the current generation of "AI" with ChatGPT for text and Dall-E for images, with Sora for video waiting in the wings. The GPT-3 foundational model is available through a chat interface for free, and GPT-4 is available for a modest fee, which includes the recently released GPT-4o. This is probably the best place to start to play with the user experience, and gradually working up to using the API for more clever things.

Microsoft saw what OpenAI did and made one of the fastest product pivots I've ever seen, spending billions of dollars to incorporate their OpenAI's models across their product line with a unified brand. Farewell Bing and Cortana, hello Copilot! It's brilliant, since you can get easy access to the GPT-4 models though Microsoft Copilot thanks to Microsoft's multibillion dollar investment and frantic launch of the new brand. They seem to be baking OpenAI technology into EVERYTHING, including Windows 11, Microsoft 365, and their Bing platform.

A number of independent platforms utilize OpenAI's models using APIs, such as perplexity.ai and magicschool.ai.

The Google Camp

Google was seemingly caught unaware by the explosive success of OpenAI and hastily launched Bard (now called Gemini). The product experience feels like they ran around to all of their AI labs screaming, "What do you have! Ship it now! Aaaaaaa!" Though Gemini started off using the Google PaLM 2 foundational model it eventually made its own eponymous model.

It's pretty remarkable how quickly Google shipped starting from nothing at the time of ChatGPT, but their chaotic approach has led to some deep blunders, such as historically implausible images and inaccurate, harmful advice. Gemini remains worth exploring, but it's showing up as an also-ran in this race.

As far as I know, Gemini is not yet the basis of any other platforms through it's APIs. Since Google is not well known for having long-term dedication to platforms or APIs, industry wariness is well founded.

The Meta Camp

Meta was also late to the game, but did not seek to directly compete with Google and OpenAI in a user-facing offer, instead promoting the open source LLaMA foundation model and emphasizing the democratization of AI and use of locally managed systems. Rather than have the models locked up in data centers at OpenAI, Microsoft or Google, practitioners could train their own models and keep total control of the training data, avoiding leaking information into the cloud or having public models incorporate private information.

Meta lets you download their models and operate on your own systems. Legendary AI professor and practitioner Yann LeCun endorses this approach, though his senior position at Meta may bias his choice!

I did notice that Facebook now provides its own "Meta AI" search, presumably based on LLaMa.

The Anthropic Camp

Antrhopic's flagship LLM platform, Claude, focuses on enterprise users. A free consumer facing chat interface is available, but you are more likely to see Claude incorporated via their API into other products, like Slack, Notion, and Zoom, as well as Quora's Poe assistant.

The Amazon Web Services Camp

Since AWS is mostly in the business of selling infrastructure, they provide a pile of parts that you can use to assemble your own LLM learning and inference pipelines.

Bedrock is the LLM managed service which lets you use whatever Foundation Model you like, whether LLaMa, Mistral, Jurassic, Claude, or even AWS's own foundation model, Titan. Waiting in the wings is AWS's Olympus foundational model.

While AWS doesn't have a chat based user experience, they provide a user-facing developer playground at partyrock.aws. (Hat tip to Chad Vavra and AWS AI solutions architect Justin Muller )

Others

While researching this article, I came across the image I used in my header (courtesy of Life Architect) which only emphasized how much more I need to learn. I don't know much about specific activity at Inflection.ai, Baidu (ERNIE), xAI (Grok) or Amazon (Olympus), but the diagram provides a teasing glimpse into the rabbit hole of LLMs.

People Smarter than Me

I've been dabbling in this space, and I've tried to direct my serious questions to folks like John Capobianco , Chad Vavra , Greg Herlein , and Cullen Jennings , all of whom have provided me insight along the way. If you found this simple article useful, let me know. If I missed anything, also let me know. Maybe I can develop interesting follow-ups based on the discussion.

Just for fun

I sometimes write about a humorous encounter with an LLM at my Quora blog, The ChatGPT Flying Circus.