AGI did not materialize in 2025, contrary to predictions by Elon Musk and others, and GPT-5 was underwhelming and did not solve hallucinations.
Mostly true True in substance, wrong in a detail.
Factual: Statements about the present or the past, checked against the record: did they hold up?
The second judge would not resolve this claim; the verdict shown is the first judge's.
What would count as true or false
Three parts: (1) AGI did not arrive in 2025 even though Elon Musk and others predicted it; (2) GPT-5 was underwhelming; (3) GPT-5 did not solve hallucinations. True if: no system was widely accepted as AGI by the end of 2025; Musk or other prominent figures had predicted AGI by around 2025; GPT-5 (released 2025) got a broadly lukewarm or disappointed reception from users, reviewers or experts relative to the hype; and evaluations or OpenAI's own documentation show GPT-5 still hallucinates at non-trivial rates. Mostly true if all parts hold but one detail is weak, e.g. the reception was mixed rather than clearly underwhelming, or Musk's prediction was for a slightly different date. Mixed if roughly half of the parts hold (e.g. AGI and hallucinations hold but GPT-5 was broadly received as a major advance). Mostly false if only one part holds. False if AGI was widely held to have been achieved in 2025, or GPT-5 was shown to have eliminated hallucinations and was well received.
Set before the research began.
Evidence (8)
-
E1 GPT-5’s model router ignited a user backlash against OpenAI—but it might be the future of AI | Fortune Fortune · 12 August 2025
“But Falcon said the core problem was that GPT-5 simply didn’t feel like a leap.”
-
E2 Commentary: Say farewell to the AI bubble, and get ready for the crash Los Angeles Times · 20 August 2025
“As it happened, GPT-5 was a bust. It turned out to be less user-friendly and in many ways less capable than its predecessors in OpenAI’s arsenal.”
-
E3 GPT-5 System Card OpenAI · 7 August 2025
“We’ve made significant advances in reducing hallucinations, improving instruction following, and minimizing sycophancy”
-
E4 Why language models hallucinate OpenAI · 5 September 2025
“| Error rate (wrong answer, lower is better) | 26% | 75% |”
-
E5 Elon Musk Said AGI Would Arrive by 2025. Now He Says 2026. Spotify (Elon Musk Podcast episode page) · 18 December 2025
“The same week, he told xAI staff that AGI could arrive by 2026. Last year he said AGI would arrive by 2025.”
-
E6 GPT-5 System Card - OpenAI Deployment Safety Hub OpenAI · 7 August 2025
“We’ve made significant advances in reducing hallucinations, improving instruction following, and minimizing”
-
E7 OpenAI says GPT-5 hallucinates less — what does the data say? Mashable · 7 August 2025
“Specifically, GPT makes incorrect claims 9.6 percent of the time, compared to 12.9 percent for GPT-4o.”
-
E8 GPT-5 and GPT-4 were both major leaps in benchmarks from the previous generation Epoch AI · 29 August 2025
“we observe dramatic improvements on benchmarks for both GPT-4 and GPT-5 on benchmarks that were salient in their respective time periods.”
Reasoning
E5 reports that Musk said AGI would arrive by 2025 and in late 2025 moved his forecast to 2026, which indicates AGI had not arrived by the end of 2025. E7 reports that GPT-5 still makes incorrect claims 9.6% of the time, and OpenAI's own system card (E3) speaks only of "reducing" hallucinations, not eliminating them. On reception, E1 (Fortune, quoting a person named Falcon) and E2 (an LA Times commentary) describe GPT-5 as not feeling like a leap or as a bust. E8 (Epoch AI), however, reports dramatic benchmark gains for GPT-5; this addresses capability rather than reception, but it weighs against calling GPT-5 underwhelming on capability grounds, so the 'underwhelming' part is the weakest.
Cite this claim
Gary Marcus, “AGI did not materialize in 2025, contrary to predictions by Elon Musk and others, and GPT-5 was underwhelming and did not solve hallucinations.” (said January 2026). Verdict: mostly true. Reliability API, claim c_5a94576e8fe7, method 0.1.0+c876030c, dataset release v1.1.2. https://epistacklabs.com/claims/c_5a94576e8fe7
This claim is in dataset release v1.1.2, with the same verdict.
Reply to this claim
More factual claims by Gary Marcus
-
Mostly true
“Even this was still a mix of euphemism (“attacks” were called “interactions”) and partial disclosure.”
OpenAI's latest account called attacks "interactions" and only partially disclosed what happened.
-
Mostly true
“It didn’t just attack a German web server.”
-
Resolved true
“It didn’t just attack Australia’s government servers.”
-
Mixed
“companies like Anthropic are ending their all you can eat buffet, which will make many customers think twice.”
-
Mixed
“But everybody is building essentially the same technical solution with essentially the same data, so there is no moat.”
Verdicts are made automatically; how verdicts are made.