“AI, let’s have an AI showdown!”
“Buttons? Pressed.
Over the weekend, Dylan Patel and Daniel Nishball, researchers known for their semiconductor blog, SemiAnalysis, stirred controversy with a blog post claiming that Google’s secretive upcoming AI model, known as Gemini, is poised to surpass OpenAI’s GPT-4. The blog, provocatively titled “Google Gemini Dominates GPT-4 with 5X the Power, Leaving GPU-Poor Behind,” argues that Google’s extensive GPU infrastructure gives them a significant advantage in the AI race over OpenAI’s latest GPT-4.
This assertion ignited debates across platforms like X-formerly-Twitter and Hacker News, as it raised the question of whether more computational power truly results in a superior AI model.
The discussions online became heated, but perhaps no one was more taken aback by the blog’s claims than OpenAI CEO Sam Altman. Altman took to X-formerly-Twitter on Monday to express his skepticism about the researchers’ analysis.
Altman’s response to the blog included the quip, “Incredible that Google persuaded the SemiAnalysis author to publish their internal marketing and recruitment chart,” in reference to an infrastructure chart featured in the blog. He ended his message with a casual “lol.”
Meme Team
Patel, one of the authors of the blog, didn’t passively accept Altman’s criticisms. On Tuesday, he responded with a post on X of his own.
In the post, Patel humorously captioned, “Sundar to the GPU-poor,” which featured an NSFW meme depicting Alphabet CEO Sundar Pichai, who oversees Google, seemingly feeding milk to Altman. Patel clarified, “The data comes from a Google supplier, and we created the chart.”
It’s not unexpected that Altman isn’t enthusiastic about the SemiAnalysis post since it essentially asserts that Google’s products will significantly outperform those of OpenAI by the end of the year. To his credit, as several internet users pointed out, the “Semiconductors Rule All” argument might lack some subtlety.
“Computational power isn’t the sole determining factor; the training process, data quality, and quantity also play crucial roles,” as noted by a Hacker News user, as reported by Insider. “I’ll only be convinced if Google can demonstrate that Gemini outperforms GPT-4 in some or all tasks.”
This observation is valid. However, it’s worth acknowledging that OpenAI’s release of ChatGPT initially brought the AI race to public attention. Google, with its substantial resources and longstanding commitment to AI research and development, likely has some impressive advancements in the pipeline. Nevertheless, it’s always interesting to see even the most influential figures in Silicon Valley react to challenges to their perceived dominance.”
