🔍 Read the full analysis: Could These 50 Signs Indicate Superintelligence? AI’s AGI And ASI Gap 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
A review published October 26 of researcher Alan D. Thompson’s 50-indicator artificial superintelligence tracker found that it lists two indicators as achieved, 33 as partially achieved and 15 as not achieved. The review says some AI mathematics results are independently supported, while claims of broad superintelligence remain unsettled and depend heavily on definitions and evidence standards.
Thompson’s tracker covers claims ranging from AI-assisted mathematical proofs to robots capable of household tasks. In an October video and the public checklist, he describes Claude Mythos Preview and GPT-6 Astra as “early ASI.” Those are Thompson’s assessments, not a consensus finding or a standard measurement of superintelligence.
The review examined 18 claims and assigned different evidence labels. It reported seven verified claims, five vendor claims, three based on a single source and three disputed or contradicted. Its verified examples were concentrated in results that can be checked against mathematical work, independent expert commentary or official records. A verified result in one task, however, does not by itself establish general superhuman capability.
The checklist’s status categories need care. Thompson’s August correction note says “partially achieved” includes documented progress such as funded programs, research infrastructure, scientific missions and enabling work. The review points to two exoplanet candidates reportedly found with Claude Code as an example of a claim assigned to an indicator about colonizing a planet beyond Earth. That classification reflects the tracker’s broad treatment of progress; it does not mean a planet has been colonized.
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
Why the score depends on definitions
The disagreement is not only about how capable current systems are; it is also about what counts as superintelligence. Thompson defines artificial general intelligence as performance at the level of an average human and artificial superintelligence as expert-human performance in practically any field. Under that wording, strong performance in expert tasks may count as evidence toward ASI.
Google DeepMind’s 2023 framework separates levels such as Competent, Expert, Virtuoso and Superhuman. Its June 2026 report describes superintelligence in more demanding terms: a system more capable than large organisations of humans. OpenAI’s charter, meanwhile, defines AGI around highly autonomous systems that outperform humans at most economically valuable work. These standards emphasize different combinations of breadth, autonomy and performance.
For readers, the distinction matters because labels such as “early ASI” can sound like a settled technical classification when they are not. A tracker can help identify claims worth checking, but its tally should not be read as a calibrated probability, a shared industry benchmark or proof that systems can reliably replace people across fields.
Products Worth Considering
Amazon Kindle Paperwhite (12th Generation, 2024), 16 GB, 7-inch Display, Jade
DJI Osmo Mobile 8 Standard Combo Phone Gimbal Stabilizer
Jackery Explorer 300 Portable Power Station, 292Wh
As an affiliate, we earn on qualifying purchases.
What the checked examples show
The review describes several AI contributions to mathematics with important qualifications. Terence Tao’s account of work on Erdős problem 728 characterized the result as “more or less autonomously” produced after feedback; the problem had been misformulated, and similar results existed. In another case, mathematician Don Knuth wrote that Claude found a construction during coached exploration, while Knuth supplied the proof. The review treats these as meaningful contributions, not wholly independent discoveries by an AI system.
It also cites an OpenAI model’s reported counterexample to a longstanding unit-distance conjecture, which the review says was supported by external mathematicians. By contrast, a separate claim about ten results on older problems, formalized in Lean for about $2,000, is labeled a vendor claim: public Lean certificates may be inspectable, but the reported figures are the company’s own and were not independently reproduced in the account.
The review uses Anthropic’s internal evaluations to illustrate limits in AI-assisted AI research. Anthropic reportedly said its best internal model scored 62.8% on a test for which the company estimated a full replacement for its researchers would require at least 85%. It also said AI made its research and development faster, “but not yet by a factor of 2.” These are company-reported measures, not independent assessments of all AI research work.
““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 on his tracker
What the tracker cannot establish
The tracker is one researcher’s framework, and its indicators mix demonstrated outputs with research programs, proposed missions and long-range outcomes. The review’s 18-claim sample is not a full audit of all 50 indicators. Its verdicts also depend on which sources and standards count as independent verification; a company’s published result may be reproducible in principle without having been reproduced by outside researchers.
The article says there is no household-robot deployment at scale, no verifiably billion-dollar AI-run company and no visible break in the US unemployment rate attributable to AI. Those observations do not settle whether AI is approaching general intelligence: adoption, economic effects and technical capability can move on different timelines. The tracker’s separate AGI countdown sits at 99%, but its final step depends on a robot completing multi-step fine assembly, such as building IKEA furniture, at average-human reliability. That figure is Thompson’s countdown, not an established measure of proximity to AGI.
It remains unclear how current systems would perform across a consistent, independently administered suite covering many expert domains, long-horizon autonomy and reliable physical tasks. The review does not establish that the models named by Thompson meet a broadly accepted definition of AGI or ASI.
What evidence would change the picture
The next useful developments are independent replications of headline research claims, clearer reporting of how much human prompting or correction was involved, and evaluations that test systems across a range of tasks rather than a single striking result. For physical-world claims, readers would need demonstrations of reliable multi-step work outside a controlled showcase, with performance compared against people doing the same tasks.
Thompson’s tracker can be followed as it changes, but its status labels should be read alongside the underlying papers, lab disclosures and expert assessments. The October review’s central test is whether each claim has evidence that outsiders can inspect—and whether the claim measures a narrow achievement or a capability broad enough to support a superintelligence label.
Key Questions
What does the 50-sign tracker report?
Thompson’s tracker lists two indicators as achieved, 33 as partially achieved and 15 as not achieved, as of October 26, 2026. He defines partial progress broadly, including funded programs and enabling work.
Does the tracker show that superintelligence has arrived?
No. It is one researcher’s checklist and interpretation, not a consensus assessment or standardized test. The review says definitions of AGI and ASI differ across researchers and labs.
Which claims did the review find strongest?
It judged several mathematical contributions to be supported by independent experts or published work. Some involved human feedback or proof-writing, so the results do not establish that an AI independently solved every part of the problem.
What remains missing from the evidence?
The review says there is no household robotics capability deployed at scale, no verifiably billion-dollar AI-run company and no broad, independently established demonstration of systems outperforming large organisations of humans across fields. Those are separate indicators, not a single test of superintelligence.
Source: ThorstenMeyerAI.com
