Marcus suspected that GPT-4's release would follow a pattern of immense buzz, then scientific scrutiny, then recognition that many problems remain.

Judge this against the period after GPT-4's release. True: GPT-4's launch drew immense initial buzz, was then followed by careful scientific evaluation (research papers, independent tests), and that work led to broad recognition that substantial problems remained, such as hallucination, reasoning errors and unreliability. Mostly true: the three stages happened, but in a slightly different order or the stages overlapped. Mixed: there was buzz, but scrutiny produced mixed conclusions, with no broad recognition that many problems remained. Mostly false: the buzz was modest, or the scrutiny mostly concluded that the major problems had been solved. False: GPT-4 was never released, or its release was followed by little buzz and the scrutiny found no significant remaining problems.

Set before the research began.

1. E1 The AI that sparked tech panic and scared world leaders heads to retirement Ars Technica · 30 April 2025

   “Its release created a wave of immense hype—and existential panic—about AI’s ability to imitate human communication and composition.”

2. E2 roundups & rapid reactions Science Media Centre · 15 March 2023

   “Although it doesn’t solve the problems of hallucinations, the OpenAI benchmarks show it is more factual by more than 40% over GPT-3.”

3. E3 GPT-4 OpenAI · 14 March 2023

   “Most importantly, it still is not fully reliable (it “hallucinates” facts and makes reasoning errors).”

4. E4 How Is ChatGPT's Behavior Changing Over Time? (Chen, Zaharia & Zou 2024) AI for Science (record summarising Chen, Zaharia & Zou, Harvard Data Science Review; arXiv:2307.09009, 2023)

   “The same service name returned substantially different behaviour within three months: GPT-4's accuracy at identifying prime versus composite numbers fell from 84% (March 2023) to 51% (June 2023).”

E1 describes GPT-4's release as creating "a wave of immense hype," which covers the buzz stage. Outside academics later measured GPT-4's behaviour in 2023 and found that its accuracy on a prime-number task fell from 84% to 51% between March and June (E4), which documents unreliability. However, the stages overlapped rather than following one another: the launch-day expert reaction already said GPT-4 doesn't solve hallucinations (E2), and OpenAI's own release page said it "still is not fully reliable" (E3). Because all three stages happened but recognition of the problems came at launch rather than only after scrutiny, the criteria point to mostly true.