The Many Flavors of AI

There is now a wide assortment of AI flavors, but the differentiation isn't merely a result of marketing. Each company has trained its model in a unique way. While the basic "Transformer" algorithmic concept is the foundation, it has been further "transformed" by how each company wants its AI to behave.

OpenAI’s ChatGPT

The first to market was OpenAI’s ChatGPT. Originally founded in 2015 with a mission to "benefit humanity," the leadership soon realized they needed billions of dollars to build massive data centers. Consequently, they transitioned to a "capped-profit" model.

If you’ve used ChatGPT, you may have noticed bifurcated responses where you were asked to select the better answer. This is an integral part of OpenAI’s training. We biological creatures were ensnared to educate the silicon ones through a process called Reinforcement Learning from Human Feedback, or the catchy acronym RLHF.

Google’s Gemini

Google, on the other hand, relied less on human consultation and more on sheer capacity. Its claim to fame is its massive context window; Gemini 1.5 can handle up to 2 million tokens at once. (And that was then!) This allows it to analyze extremely long, complex legal documents so us poor humans can understand what they actually mean. In addition, Google utilized all forms of media (text, images, video, and audio) from the start, so it sees our world as seamlessly interconnected.

Anthropic’s Claude

Then there is Anthropic’s Claude, the AI with my favorite name. Its distinction lies in being trained on a "constitution," a hybrid of the UN Declaration of Human Rights, Apple’s terms of service, and other assorted principles. Because it bypasses the unpredictable passions of humans, it tends to be verbose and cautious, even "preachy" by some accounts. This is Anthropic’s "neutral" business model: not too hot, not too cold, but just right.

Microsoft’s CoPilot

Microsoft liked OpenAI’s model so much they bought the company! While they didn't actually buy the company, they heavily invested in the model, integrating it across all Microsoft software. This allows the AI to work alongside you by accessing the information you're already handling. Its value lies not just in chatting, but in doing. This strategy wasn't lost on its competitors. Google just this January launched Personal Intelligence which allows access Gmail, Google Drive, and YouTube.

Elon Musk’s Grok

Grok prides itself on being the "anti-woke" alternative. Elon Musk, a co-founder of OpenAI, wanted a product that would answer questions other bots refuse, confirming his stance as a free-speech absolutist. Since it has exclusive access to X (formerly Twitter), it is highly competitive in delivering real-time analysis of global events.

Meta’s Llama

While OpenAI and Google kept their models proprietary, Meta took a radically different approach. Arguing that AI should be a shared scientific endeavor, they championed an "open-source" strategy, a path OpenAI originally took before abandoning it. This allowed thousands of niche variants like Alpaca and Vicuna to flourish. Meta’s strategy is clear: commoditize AI so no single competitor can dominate and force others to conform to its proprietary rules.

As for me, I am using Gemini. I haven’t had the time to test every other flavor, so I can’t definitively say it’s the "best." However, its Veo 3.1 video generation is of astounding high quality, and I haven’t even scratched the surface of its capabilities. I remain curious about the evolution of these tools and look forward to hearing from those who use the other services.