The recent claim that GPT-4 would possess 100 trillion parameters has gone viral on social media, and I must admit that I played a role in its propagation.

To illustrate, consider the two visuals shared on Twitter that have collectively garnered around 5 million views. Variants of these images are also circulating on LinkedIn, Reddit, and other platforms, each presenting a compelling graph comparing GPT-3 and GPT-4. These posts often come with emotionally charged remarks like “this is a frightening visual” or “[GPT-4] will make ChatGPT seem trivial.” The assertions made are often presented without any evidence, leading to a false sense of certainty among millions who first encounter GPT-4 through these posts.

This creates a troubling scenario: individuals are left with inflated expectations based on misleading information, only to be disillusioned when the reality of GPT-4 does not meet these exaggerated claims. Such rampant misinformation could contribute to an "AI winter," a term used to describe a downturn in AI development and investment. While I believe this extreme outcome is unlikely, the risk is heightened by the spread of such falsehoods.

This piece serves as both a cautionary note about the rapid dissemination of misinformation online and a personal reflection on how sharing knowledge without due diligence can have unforeseen negative consequences.

The genesis of this misinformation can be traced back to an article by Will Knight published in Wired in August 2021, which reported on a conversation with Andrew Feldman, the CEO of Cerebras Systems. Feldman claimed that GPT-4 would have around 100 trillion parameters, which I later echoed in my own article published in September 2021. At that time, it felt like I was sharing cutting-edge information, but in retrospect, I failed to scrutinize the speculative nature of the claim adequately.

Initially, I framed the information as a definitive statement rather than recognizing its hypothetical nature. Feldman’s assertion was clearly uncertain; he used phrases like “about” and “won’t be ready for several years,” indicating a lack of concrete information. Yet, I presented it with a sense of finality that contributed to the hype.

When new information emerged, including a denial of the 100 trillion parameter claim from Sam Altman during a Q&A session, I eventually added a disclaimer to my original article to correct the misinformation. However, by that time, the misleading visuals had already spread widely, often overshadowing the clarifications I had made.

The arrival of ChatGPT only intensified the situation, as discussions about GPT-4 became more prevalent, with some insisting it would be massive while others noted the debunked claims. The allure of sensational topics like AI often overshadows the importance of accuracy, leading to a cycle of misinformation that is difficult to break.

Reflecting on my role in this narrative, I recognize the importance of balancing engaging content with responsible reporting. While it’s crucial to address the concerns and excitement surrounding AI, it’s equally important to ensure that the information shared is framed appropriately.

I take full responsibility for contributing to the hype surrounding the “100 trillion parameters” claim. Moving forward, I am committed to presenting information more thoughtfully, ensuring that my sources are reliable and that the likelihood of their claims is accurately represented.

Subscribe to The Algorithmic Bridge for insights on AI that matter to your life. Support my work on Medium by becoming a member using my referral link here! :)