Upgrade your account to a Community Supporter account and remove most of the site ads.

Reply to thread

Message: <blockquote data-quote="inf0squit0" data-source="post: 9768835" data-attributes="member: 7054116">There exist some methodological issues with the reported ChatGPT user counts - very little was done to detect logged-out users "double dipping" by using multiple devices, clearing their browser cache, etc, and there's also no accounting for how engaged any of those users remained. This data counts every person who so much as dropped one prompt in the text box just to see what the hype is about. I would also not be surprised if dozens of "users" are being generated by every illicit ChatGPT user trying to trick it into writing ransomware for them, and when they fail, they clear their browser cache, change their VPN exit node, and try again.Plus trusting anything coming out of OpenAI is a tough sell as it is, they have little reason to be honest in their reporting since they're not publicly traded. These user counts are almost certainly also being inflated by bundling subscriptions with other things, like <a href="https://laist.com/news/education/chatgpt-california-state-university-csu-ai-deal" target="_blank">university enrollment</a>. Plus there are reports from ChatGPT users that they're being offered deep discounts that are, measured against the cost of operating these models (a cost which doesn't seem to be going down, unlike past success stories with scalable IT products) likely unsustainable in the long run and may very well be helping to cover up user churn that is far worse than it looks from the outside. There's a very real possibility that AI companies are more like WeWork than Uber and OpenAI's self-reported success and growth is a shell game to disguise the realization that their novelty shovelware-generator isn't actually useful or reliable for creating products worth selling (or buying).There's very little objective information that supports the idea that LLMs are being implemented in ways that are actually useful. <a href="https://www.newscientist.com/article/2479545-ai-hallucinations-are-getting-worse-and-theyre-here-to-stay/" target="_blank">Hallucinations are getting worse</a>. <a href="https://www.futurehouse.org/research-announcements/hle-exam" target="_blank">The benchmarks showing AI success have profound issues. </a><a href="https://blog.kilocode.ai/p/future-ai-spend-100k-per-dev?ref=wheresyoured.at" target="_blank">AI keeps getting more expensive per user to operate</a>. <a href="https://fortune.com/2025/09/23/ai-boom-unsustainable-tech-spending-parabolic-deutsche-bank/" target="_blank">Big finance is getting cold feet</a>. <a href="https://www.theregister.com/2025/06/16/salesforce_llm_agents_benchmark" target="_blank">AI agents can't reliably complete basic tasks, much less complex ones.</a>It's not like LLMs solve zero problems, but IMO the major issue with the AI/LLM discourse is how frequently the number of problems AI can reliably, efficiently solve is vastly overestimated. And that many of the problems LLMs do solve aren't problems reasonable people wanted solved - like how to more efficiently commit fraud, develop ransomware that avoids anti-malware software, clog the public Internet with slop text, etc.</blockquote>

Verification