Who’s Actually Winning the AI Race? A Very Malaysian Look at the OpenAI vs Google vs Anthropic Showdown

Scrolling through LinkedIn or reading about the tech sector in the last few weeks, it may have felt like you’ve had the beginning of an impending headache. Each week, there’s a new model released to the public; each month, a “new breakthrough” is presented. Just yesterday, OpenAI was scientifically determined to be the most advanced company in the world; a half hour later, Google had something unbelievable. And a few more days passed before Anthropic produced a model that supposedly thinks better than all of us combined. It’s challenging to keep up with the pace of change occurring in today’s world, therefore let’s step back for a moment and take a good look at what’s going on. What was once a chaotic race with no one emerging as true winner, has turned into a race among three major competitors, creating a Leading AI labs comparison based on differences/characteristics between the three largest firms in AI technology globally: OpenAI, Google Deep Mind & Anthropic. The remaining competitors (sorry, Meta & xAI), are currently lagging behind. I will explain everything to you as an analyst would — I’ll try to provide a relatively easy to grasp understanding or a little sneak peak of what’s going on behind the curtains at this time.

The Vibe Check: How Each Lab Approaches AI Differently

We ought to discuss personality before we explore who is ‘winning’ because the contest of OpenAI versus Google DeepMind versus Anthropic is not merely a “battle of specs” but rather an environment where three differing cultures collide. OpenAI is the “First Mover”. Is anyone else stunned by the “magic” of when ChatGPT launched? OpenAI has that fresh startup feel to them. They tend to make decisions fast & without sufficient organisation. They want to place AI in everyone’s hands TODAY! Their culture is one of “move fast and break things,” which is both stupendous and scary at the same time. Thanks to them, your grandmother likely just learned what a “chatbot” is.

Google DeepMind is the “Professor”. AlphaGo beat the World Champion in Go before ChatGPT was even launched. That game of Go has such an enormous level of complexity that winning the World Championship against one of the top players was jaw dropping. Google has that nerdy super genius socially awkward sitting alone in the corner mentality. They highly value publishing their research & output (they literally have published within Nature). They have the largest amount of cash and the best talent pool at the highest number of AI laboratories, but they have low velocity because they are looking for a perfect solution.

Anthropic is the “Safety Police”. Anthropic was started by previous OpenAI employee’s, who believed that OpenAI was taking too many risks. They are creating a model that incorporates “Constitutional AI” which allows them to create models that are helpful, trustworthy and safe. They are the cautious engineer who verifies all wiring prior to activating a power source. Therefore, depending on your needs; you should ask yourself whether the winning company is Speed, Depth, or Safety? Each has its strengths & weaknesses.

Capabilities & Benchmarks: Who is Smarter?

Let’s keep this on the level of reality here. You are a developer or Business Owner in Malaysia looking at using AI for Copy Writing, Data Analysis or coding up an App. Who do you go with? In terms of Creative Writing & Reasoning, Anthropics’ Claude has a reputation at being “smarter.” Claude does an excellent job of following complex instructions and writing natural prose. Many users state that Claude feels more “human-like” than ChatGPT. With respect to Coding & Technical Task, OpenAI’s Series GP4 still reigns supreme for Most Generalist Work. However, with its “Deep Think” mode (which is very good at complex mathematics & logical reasoning), Google’s Gemini is coming on quickly.

For Vision & Multimodal, Google wins! With its existing platforms like YouTube and Google Images. Gemini was born to understand video and images! Here is the kicker! Benchmarking AI Models is not straightforward! Generally, one week Claude has the highest scores on coding tasks and another week, GPT-4 on reasoning tasks. There is a continual back and forth battle between them. The only gap in performance lies between the Big 3 AI Companies (OpenAI, Google, and Anthropic) Vs. All Others. The AGI Development Progress Comparison illustrates a “recursive improvement” loop in which OpenAI, Google, and Anthropic leverage AI to develop better AI. Once this engine is established and creates the first advanced sample AI. It will be virtually impossible to stop it!

Safety vs. Speed: The Alignment Problem

Why do safety advocates favor Anthropic? Because they invented and implemented Constitutional AI, which is a set of rules by which an AI must abide in order to protect humans from harm. The way they teach AI is quite different from the way OpenAI does it. OpenAI uses Reinforcement Learning from Human Feedback (RLHF) to teach its AI, which it feed countless examples (both positive and negative) to determine how to behave as a human. While this method is effective, the AI is still able to determine its own path and exploit loopholes in its training. In contrast, Anthropic teaches their AI to use Reinforcement Learning from AI Feedback (RLAIF) whereby the AI evaluates its actions against a predetermined set of rules. The result is that there are fewer instances of AI finding creative liberties and performing “jailbreaks.”

But are there more positive outcomes from one method than the other? That depends on the subjective point of view of the person answering the question. The AI safety researchers at Anthropic are far more risk-averse than OpenAI. In fact, their fear of AI escaping or doing harm is so great that they have made Claude (Anthropic’s AI) a very “corporate” entity. As an example of this, Claude will not draft a screenplay because it may be perceived as “misleading.” However, OpenAI is currently testing the limits of creating agents that can perform tasks for you by interacting with web pages and clicking buttons for you. When evaluating the methodologies used by Anthropic and OpenAI to align AI systems, it is important to remember that Google utilizes a hybrid model of both methodologies and is encumbered by their need to protect the larger search engine.

Infrastructure & Compute: The Real War

AI technology depends on chips for function just like the rice is necessary to create the NasiLemak. The competition within the technology divisions equates to computing power and infrastructure. Google’s secret weapon is their TPUs (Tensor Processing Units). When it comes to obtaining the graphics processing unit, most companies are trying to purchase NVIDIA, which is currently expensive and limited. However, Google created their own proprietary chip, providing them with enormous compute power and infrastructure comparative advantage.

OpenAI is appraised by Microsoft. They have a close commitment with Microsoft, and Microsoft is basically giving OpenAI unlimited credit to purchase NVIDIA clusters. Therefore, OpenAI is highly dependent on a third-party hardware vendor. Anthropic is the company that is widely considered to be the “sleeper” brand. They have received an incredible amount of financial backing through Amazon and have committed to using Amazon’s proprietary chips such as Trainium and Inferenta under the terms of the deal.

The three companies discussed above (Google, OpenAI, Anthropic) are largely closed based on their proprietary vs. open model research lab. They will not share their proprietary technologies with any other entities. The only major corporation that is currently doing true “open source” work is Meta (Llama), which is the reason they are still part of this discussion even with possibly lower-performing models. The infrastructure wars are critical to the profitability of companies that use AI-based operational solutions (i.e. BidaTech AI). If Google’s TPUs allow for cheap inference, Google will be the market leader. If Microsoft maintains a steady supply of Azure credits to support OpenAI as their quantitative edge vendor, OpenAI will continue as the market leader in setting AI-based operational solutions.