The AI Bubble Burst
When will the market figure it out?
The chart we published yesterday showed declining ChatGPT usage. That is effectively the AI bubble bursting.
The “why” is related to a number of reasons, in no particular order:
The more I’ve used LLMs, the more I am confronted with their limitations. LLMs “guess”, they don’t “know” facts. LLMs “guess” answers based on ingested information and the probability distribution of potential answers, not based upon learned experience as humans do. LLMs often fail to quickly course correct when commons sense suggests an alternative approach to problem solving. Language models can’t be trusted to correctly read and interpret data that has been fed to them, as a result their output is often factually incorrect. This means time saved on the production side is often lost on the quality control side, which limits ROI.
There are an abundance of open source LLMs. Open source LLMs are far cheaper, allow for downloading to local machines (privacy), and enable customization of weights. Privacy and customization matter to researchers.
Claude is superior to GPT for coding, especially for front-end work.
Many enterprises that experimented with LLMs realized the ROI was not in-line with expectations (that’s a separate note). Thus, less repeat usage.
LLM users who preferred the Google search box now have Gemini included in Google search. Gemini 2.5 Pro is quite good, especially for every day use. Gemini 3.0 will be released in December and will likely be superior to GPT 5 and Claude Sonnet 4.5 in every way. Google will be the LLM king when all is said and done.
Try T2D Pulse:
