🔍 Read the full analysis: 50 Signs Of Superintelligence: How To Judge AI’s Progress on ThorstenMeyerAI.com
Get the latest gadgets delivered free with Prime
- Fast, free delivery on millions of items
- Prime Video, Amazon Music and more included
- Member-only deals all year
TL;DR
Thorsten Meyer’s October 26 review checks 18 claims from AI consultant Alan D. Thompson’s 50-indicator superintelligence tracker. It classifies seven as verified, five as vendor claims, three as supported by a single source and three as disputed or contradicted; the tracker itself marks only two of its 50 indicators achieved.
Thorsten Meyer published a review on October 26 checking 18 prominent claims in AI consultant Alan D. Thompson’s list of 50 signs that artificial superintelligence has arrived. Meyer reports that seven claims were verified, five were vendor claims, three rested on a single source and three were disputed or contradicted, underscoring how different the evidence can be from the tracker’s broad progress labels.
Thompson’s public tracker, last updated October 26, lists two indicators as achieved, 33 as partially achieved and 15 as not achieved. In an August correction note, Thompson said “partially achieved” covers documented progress that can include funded programs, research infrastructure, formal scientific missions and other enabling work. The category is therefore not a measure that an outcome is nearly complete.
Meyer says he selected 18 strong claims from Thompson’s October video and tracker and traced them to primary sources, including research papers, lab statements, a researcher’s account or official filings. The evidence was strongest for results that could be checked directly, such as mathematical work, a validated physics calculation and regulatory documents. Other claims relied on companies’ own reporting or had limited independent support.
The review highlights AI contributions to mathematics while retaining caveats about how they were achieved. Terence Tao described one result as reached “more or less autonomously,” but said the process involved feedback and tools including Aristotle and GPT-5.2 Pro; the problem had also been misformulated, and similar results existed. Donald Knuth documented Claude finding a construction after coached exploration, while Knuth wrote the proof. Meyer also reports external mathematicians verified an OpenAI model’s result concerning a unit-distance conjecture.
Whose superintelligence?
One word, very different bars
Dr Alan D. Thompson's ASI starts where Google DeepMind's scale puts "Expert AGI". The same model can be "early ASI" on one scale and a strong but narrow AGI candidate on another.
- Median human50th percentileThompson: AGIa machine at the level of an average (median) humanDeepMind: Competent AGIat least the 50th percentile of skilled adults
- Expert90th percentileThompson: ASIexpert-human level in practically any fieldDeepMind: Expert AGIat least the 90th percentile of skilled adults
- Virtuoso99th percentile—DeepMind: Virtuoso AGIat least the 99th percentile of skilled adults
- Beyond every human100%—DeepMind: Superhuman (Level 5)outperforms 100% of humans
- Beyond organisations2026 report—DeepMind: ASImore capable than large organisations of humans
Thompson’s tracker status as of 26 October 2026. DeepMind rows: “Levels of AGI” (2023) and, for the last row, “From AGI to ASI” (June 2026). OpenAI’s Charter: AGI = “highly autonomous systems that outperform humans at most economically valuable work”.
Sources: LifeArchitect.ai ASI checklist (26 Oct 2026), AGI countdown, Morris et al., Levels of AGI (2023), Genewein et al., From AGI to ASI (2026), OpenAI Charter.
Powered by Thorsten Meyer AI
What the Evidence Supports
The review matters because “superintelligence” is not being used consistently, and a checklist can make unlike developments appear comparable. Thompson defines ASI as a machine performing at expert-human level in practically any field. Google DeepMind’s 2023 scale calls a 90th-percentile level “Expert” and distinguishes it from “Superhuman,” which means outperforming all humans. DeepMind’s June 2026 report uses a more demanding description: systems more capable than large organisations of humans.
Those definitions change how readers should interpret claims that a model is “early ASI.” A system producing an expert-level result in one domain may be evidence of substantial capability without showing broad, autonomous superiority. OpenAI’s charter, cited by Meyer, defines AGI around highly autonomous systems outperforming humans at most economically valuable work. Capability, breadth and autonomy are related but distinct tests.
For workers, businesses and policymakers, the distinction affects expectations about economic change. Meyer’s review says it found no visible break in the U.S. unemployment rate attributable to AI and no verifiable AI-run company worth a billion dollars. Those observations do not settle AI’s future economic effects, but they caution against treating a fast-moving research result as proof that broad social or workplace transformation has already occurred.
Products Worth Considering
Amazon Kindle Paperwhite (12th Generation, 2024), 16 GB, 7-inch Display, Jade
Sony WH-1000XM4C Wireless Noise-Cancelling Headphones, Black
Jackery Explorer 300 Portable Power Station, 292Wh
As an affiliate, we earn on qualifying purchases.
How Thompson Defines Progress
Thompson, an Australian AI consultant who publishes The Memo newsletter, maintains a checklist spanning claims from recursive code self-optimization to far-future possibilities. His October video, “ASI: A better world – 2026,” presents a montage of developments from January through October. He identifies Claude Mythos Preview, released in February according to the source material, and GPT-6 Astra, which he says appeared in September, as “early ASI.” These are Thompson’s characterizations, not a consensus classification established by the review.
The tracker’s status labels need to be read alongside its definitions. Thompson’s August note says orange, or partial, includes enabling activity rather than only near-completion. One example cited in the source is two exoplanet candidates reportedly found with Claude Code being placed under an indicator about colonizing a planet beyond Earth. That illustrates how a tracker category can connect an early research development to a much larger eventual outcome.
Thompson also keeps a separate “conservative countdown to AGI,” which the source says stood at 99%. He says the remaining point depends on a robot completing multi-step fine assembly, such as assembling furniture, at average-human reliability. This is a separate estimate from the 50-indicator ASI tracker, not an independently measured probability that AGI is 99% complete.
“documented material progress toward an indicator, including funded programs, research infrastructure, formal scientific missions, and enabling work”
— Alan D. Thompson, in an August correction note about the tracker
Where the Claims Remain Unsettled
The review’s findings apply to the 18 claims Meyer selected, not an independent audit of all 50 indicators. The source material does not provide enough detail to assess every claim or reproduce the full evidence behind each verdict. The labels also capture different kinds of support: a result verified by an outside expert is not equivalent to a company’s own performance report.
Some prominent claims remain dependent on vendor reporting. Meyer classifies five of the 18 as vendor claims, including an OpenAI claim that GPT-6 Astra produced ten results on decade-old problems for about $2,000, with results formalized in Lean. Public Lean certificates are mentioned, but the source says human preparation was involved. The materials provided do not establish that the overall claim was independently reproduced or clarify all costs and human contributions.
It is also unclear how well performance on selected proofs or scientific tasks predicts reliable work across fields, sustained autonomy or real-world economic impact. The source gives no independent measurement demonstrating that current models meet a shared definition of ASI. Its account of unemployment and company valuations likewise does not establish that AI has had no effects; it says only that the review found no visible unemployment break or verifiable billion-dollar AI-run company.
What Future Checks Must Show
The immediate next step is to evaluate new claims against their primary sources and separate independently checked results from company-reported measures. For mathematical work, that means tracking whether proofs and problem statements withstand expert review and documenting the roles of models, tools and human feedback. For commercial claims, readers need clear accounting of costs, human input and the degree to which results can be reproduced.
Progress on the tracker’s unfinished indicators will also depend on what Thompson counts as evidence. His separate AGI countdown points to a practical test: a robot completing multi-step fine assembly at average-human reliability. The source material does not give a date for that milestone, and the tracker’s partial category permits enabling work well before it is achieved.
For now, the status remains mixed: some AI contributions have outside support, while other high-profile claims remain vendor reports, single-source accounts or disputed. Further independent replication and clearer definitions of AGI and ASI would make future updates easier to compare.
Key Questions
What did Meyer’s review find?
Among 18 selected claims, Meyer classified seven as verified, five as vendor claims, three as based on a single source and three as disputed or contradicted.
How many of Thompson’s 50 indicators are marked achieved?
The tracker showed two achieved, 33 partially achieved and 15 not achieved as of October 26, 2026. Thompson’s definition of partial progress includes enabling work, so it does not mean an indicator is nearly complete.
Does the review establish that AI has reached superintelligence?
No. It reports evidence for particular capabilities but also notes that definitions of ASI differ and that several claims lack independent confirmation. It does not establish that current systems meet a shared standard for superintelligence.
What is the main caveat about AI’s math results?
Some results involved human feedback, coaching or proof-writing, and at least one problem had been misformulated or had similar existing results. The source distinguishes those contributions from fully independent, general mathematical ability.
Source: ThorstenMeyerAI.com
