Ever since OpenAI released chatGPT 3.0 to the public, the internet was taken by storm, leading to a surge in A.I. adoption. Many companies joined the race to develop their own models to compete in this emerging field but two stand out: OpenAI, obviously, and Google.
But how do these two companies compare, and, since you read the title, what makes Google fall short? In this article, I’ll go through some of their past developments and contributions and talk about current shortcomings.
In the Past
Before chatGPT gained attention for potentially changing searches forever, by providing customized answers without any need for clicking a website with multiple Ads, Google had already made significant strides in integrating AI into its products. It applied machine learning in various areas, including YouTube for content moderation and recommendations, Google Assistant, Android’s keyboard, Maps, and more. These efforts aimed to automate processes and enhance efficiency.
While Google’s approach to AI implementation might have been less sophisticated than current modern models, it represented a substantial leap in complexity and technological capabilities. Google contributed to AI development through open research papers, substantial investments, and the early development of TPUs (Tensor Processing Units). Sundar Pichai, the CEO in 2016, even stated “Google, whose name had become synonymous with search, would now be an “AI-first” company”. Why didn’t this transition succeed as expected?
Failure
In all fairness, signs of Google’s AI limitations were evident. Throughout its history, Google had not produced any advanced machine learning models. In the context of this article, YouTube serves as a glaring example of Google’s struggles with AI, despite its access to vast amounts of data and pioneering in the area.
If you read the homepage of this blog, you know I have been a video editor for multiple youtubers for at least two years now. As such, I have first hand experience with how Youtube works behind the scenes so allow me to warn you: it smells. Youtube has billions upon billions of videos with millions of hours of content in all formats, serving as a massive dataset of pretty much anything audio-visual, and Google has had years to develop artificial intelligence for checking what content said videos have and whether or not they break the T.O.S.
And yet, YouTube’s algorithm frequently mislabels videos as “unsuitable for all viewers,” even for innocent content like a bunny avatar laughing or a non-explicit furry character taking a bath. Furthermore, certain sound effects, like bass boosting, are inaccurately identified as “offensive language.” more often than not. Google’s AI also tends to misinterpret substances resembling white liquid (e.g., milk) in a way that you can likely guess.
Moreover, Google possesses a vast and well-organized text database used for search, and yet it failed to create an effective chatbot, despite having double the development time compared to OpenAI. Their employees have described the chatbot as “worse than useless.”.
Gemini
Recently, google announced its newest attempt at making an all-purpose A.I.: Gemini. In the presentation, they showed it recognizing objects in real time within a video, among other things.
This, as is known now, was faked. Google faked making an AI that can recognize things in a video as its playing. The same company that, again, owns Youtube, the biggest archive of videos of all kinds and formats ever created, and some of the biggest, most powerful datacenters in the world, couldn’t make a model capable of properly analizing videos, and yet has tried implementing that for commercial reasons within Youtube, to cheap out on human resources.
Update: Gemini is a complete disaster. It was recently shown to even refuse to code some C++ and/or other languages if the person asking is underage. It’s image generation was put on pause due to producing historically inaccurate images.
Conclusion
Following the popularity of chatGPT, numerous companies sought to develop their machine learning models. While this task is challenging, particularly for smaller entities, it raises questions when a company as well-funded and experienced as Google fails so remarkably.
While rushing the release may have played a role, the complete explanation remains a mystery. However, it is evident that Google’s performance in AI did not match that of other companies of same size. Despite the setbacks, witnessing a technological monopoly lose ground, even if only slightly, can be seen as a positive development.
