I have two independent primary sources (a government-hosted peer-reviewed journal article on NIH's PMC, and a peer-reviewed journal article in Nature's *npj Digital Medicine*), plus OpenAI's own primary release. That's sufficient. VERDICT: UNVERIFIABLE **The claim:** A TikTok video asserts "ChatGPT 5 might be the greatest invention of all time." This is a subjective superlative ("greatest … of all time") hedged with "might be" — it is an opinion about relative historical significance, not a testable factual proposition. No authoritative body ranks inventions, and there is no measurable metric for "greatest invention of all time." So the statement cannot be confirmed or refuted as a fact; what *can* be checked is whether the evidence about GPT-5 supports the hype framing. **What the primary evidence actually shows (claim-linked findings):** - **GPT-5 is a real, documented advance — but a context-dependent one.** A peer-reviewed letter in the journal *Malaysian Family Physician*, hosted on the U.S. National Library of Medicine's PMC (a government domain), reports that GPT-5's hallucination rate fell to **9.6% (from 12.9% for GPT-4o)** with web access, and its reasoning variant reached **4.5%**, producing **44% fewer responses containing at least one major factual error** versus GPT-4o. Critically, the same source notes that **without internet access on fact-seeking tasks, GPT-5's hallucination rate rose to 47%**, and that "human oversight remains indispensable, particularly in high-stakes domains such as medicine." ([PMC12701941 – Polat et al., *Malays Fam Physician*, 2025](https://pmc.ncbi.nlm.nih.gov/articles/PMC12701941/)) - **Independent re-testing found persistent risks that undercut "greatest" framing.** A peer-reviewed study in *npj Digital Medicine* (Nature portfolio) re-ran validated clinical pipelines on GPT-5 and found it "showed no measurable improvement over GPT-4o in sociodemographic-linked decision variation," with several LGBTQIA+ groups flagged for mental-health screening in **100% of cases** despite identical clinical details, and **adversarial hallucination rates that were higher than GPT-4o (65% vs 53%)**. ([Omar et al., *npj Digital Medicine* 9:282 (2026)](https://www.nature.com/articles/s41746-026-02584-8)) - **OpenAI itself frames GPT-5 as "our smartest, fastest, most useful model yet"** — a claim about being its best model to date (with benchmark figures such as 94.6% on AIME 2025, 74.9% on SWE-bench Verified), **not** a claim of being the greatest invention in human history. ([OpenAI – "Introducing GPT-5," Aug 7, 2025](https://openai.com/index/introducing-gpt-5/)) **Why the verdict is UNVERIFIABLE rather than TRUE/FALSE:** - The core phrase is a value judgment about historical significance; it has no falsifiable standard or authoritative ranking. - Even if reframed as a factual capability claim, the independent research shows a **mixed** picture — genuine measurable gains on some tasks, alongside unchanged bias and, in one adversarial test, *worse* hallucination behavior than the prior model — so "greatest … of all time" is neither established nor disprovable by evidence. **Bottom line:** The video's statement is an opinion/hype claim, not a verifiable fact — hence **UNVERIFIABLE**. Evidence supports GPT-5 being an incremental but notable advance; it does **not** support (or refute) the superlative "greatest invention of all time." **Primary sources cited (opened):** 1. https://pmc.ncbi.nlm.nih.gov/articles/PMC12701941/ — government domain (NIH/NCBI), peer-reviewed article. 2. https://www.nature.com/articles/s41746-026-02584-8 — peer-reviewed journal (*npj Digital Medicine*). 3. https://openai.com/index/introducing-gpt-5/ — developer's own primary announcement (context on what GPT-5 is claimed to be).