ChatGPT Has Been Slumping on Google Trends Since April -- And Outranked by Leading Metaverse Platforms Roblox, Minecraft & Fortnite

According to Google Trends -- which is actually more or less the collective global brain -- interest in ChatGPT peaked last April, for a brief time, and has been trending down ever since -- even though there have been numerous updates and expansions to the program last May, June, and July.

More key for this blog, ChatGPT only briefly outpaced leading metaverse platforms Roblox, Minecraft and Fortnite on Google Trends in April. Before and since, however, it's trailed behind, usually by quite a lot.

Why? For one thing, platforms comprised of live, active, masses of creative people will always be more popular than an automated service. For another, ChatGPT like all other Large Language Models are by definition mediocre content generators:

When you ask ChatGPT a question, you get a response that fits the patterns of the probability distribution of language that the model has seen before. It does not reflect knowledge, facts, or insights.

And to make this even more fun, in that compression of language patterns, we also magnify the bias in the underlying language. This is how we end up with models that are even more problematic than what you find on the web. ChatGPT specifically does use human-corrected judgements to reduce the worst of this behavior, but those efforts are far from foolproof, and the model still reflects the biases of the humans doing the correcting.

Finally, because of the way these models are designed, they are at best, a representation of the average language used on the internet. By design, ChatGPT aspires to be the most mediocre web content you can imagine.

More here. Click to embiggenate the full Google Trends report.

Comments

[John]

And not only is the content 'mediocre', but it unavoidably gets MORE mediocre and fuzzy over time, by its very nature (as explained by super SF writer Ted Chiang): https://www.newyorker.com/tech/annals-of-technology/chatgpt-is-a-blurry-jpeg-of-the-web

[Nadeja]

Another explanation is that, in summer, there is more interest in games and less in homework-aid because students are on holidays.

If we zoom out and look at the 5 year graph to get the bigger picture, most of the spikes in Google Trends for Minecraft and Roblox happen in summer and Christmas; the effects of the SARS-CoV-2 pandemic are also clearly visible. Fornite is on a slow downtrend since years. There are multiple factors for these trends, tough (also note that the graph is the USA only, while by choosing "Worldwide" the trending down of ChatGTP on Google search is less pronounced).
Sure, another factor may be that the initial hype on ChatGPT may be fading out and some people couldn't find any better use, especially those on the free version that has no Internet access and is based on the outdated GPT-3.5 and they are unaware of the alternatives (here is a failure by Microsoft in promoting their Bing Chat, for example, for several reasons).
Given that, I think it may be too soon to conclude anything now: the interest could fade out or the trend invert again by October. By then or December, though, maybe Google Gemini would be the next thing (unless Google flops again as they did with Google Bard).

What worries me most, short-term, is the use of LLMs by propaganda bots and generated pictures and videos that may affect the 2024 election in the USA. They are surely not less interested than before.

[Iggy 1.0]

Hamlet, why continue to compare it to metaverses? That is like comparing sales of bicycles to motor vehicles, because they all have tires.

And did you consider / measure ChatGPT usage as it is now rolled into Bing? Numbers will grow there. A better comparison would be as a percentage of queries, as compared to traditional search engines. I bet Microsoft will have those data for Bing with and without AI.

In higher ed, generative AI is huge. It is now incorporated into Grammarly and Google Workspaces. It feels different from our short-lived infatuation with SL, which had a clunky UI, ran poorly on student laptops, and felt “creepy” (their favored term than) to many undergrads and creepy and pointless to many administrators.

As for the quality of the AI output? There you make a more interesting point. Yet even there, as my six months of testing indicate, the ChatGPT 3.5 made enormous progress in stylistic variation, transitions between points, and quality of arguments. ChatGPT 4 is even better.

Sure, it’s still a synthesis but better than what most of my first-years produce. The makers have turned off the ability to quote directly from any works, even those in public domain, out of fear of copyright lawyers (fair use laws vary, globally).

I suspect that unlike SL, generative AI will scale. Keep revisting this topic, but please stop comparing it to Metaverses: Different applications completely.