🔍 Read the full analysis: Assessing AI’s Next Steps: 50 Signs Of Superintelligence Checked on ThorstenMeyerAI.com
Get the little things that make your day 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 by ThorstenMeyerAI.com on October 26 examined 18 prominent claims from researcher Alan D. Thompson’s 50-sign artificial superintelligence tracker. It classified seven as verified, five as vendor claims, three as based on a single source and three as disputed or contradicted; the tracker itself lists two indicators achieved, 33 partially achieved and 15 not achieved.
A review published October 26 by ThorstenMeyerAI.com checked 18 prominent claims from researcher Alan D. Thompson’s 50-sign artificial superintelligence tracker, finding seven verified, five supported only by company claims, three based on a single source and three disputed or contradicted. Thompson’s tracker marks two of its 50 indicators achieved, 33 partially achieved and 15 not achieved, but the review says those labels should not be read as a measured countdown to superintelligence.
The review traced selected claims to papers, lab announcements, researcher accounts and regulatory filings, and classified their evidence using five labels: verified, vendor claim, single source, disputed or contradicted, and not found. Its overall conclusion was that results are strongest when they can be checked independently, including mathematical work, a physics calculation and regulatory filings. The supplied source does not provide the full verdict table for all 18 claims, so the individual breakdown beyond the named examples cannot be reconstructed here.
Several mathematics examples received favorable assessments with qualifications. Mathematician Terence Tao described GPT-5.2’s work on an Erdős problem as “more or less autonomously,” while noting it followed feedback, used specific tools and addressed a problem that had been misformulated, with similar results already available. Donald Knuth wrote that Claude found a construction for odd sizes in an open-problem investigation after coached exploration; Knuth supplied the proof, and later GPT models helped with the even case. The review also says external mathematicians verified an OpenAI model’s disproof of a 1946 Erdős conjecture.
Other claims carry less independent confirmation. The review classifies an OpenAI claim about GPT-6 Astra producing ten results on decade-old problems for about $2,000 as a vendor claim, while noting that Lean certificates were public and humans prepared them. On AI-assisted research and development, Anthropic’s own account put its best internal model at 62.8% on a test, compared with an estimated 85% threshold for fully replacing its researchers. Anthropic also said its research and development was faster with AI, “but not yet by a factor of 2.”
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
Evidence Varies Across AI Milestones
The review matters because labels such as “early ASI” can sound like a settled scientific finding, even when the underlying evidence ranges from independently checked results to company-reported performance. A tracker that counts funded programmes and enabling work as partial progress is recording activity toward an indicator, not necessarily demonstrating that the capability itself has arrived.
The source also identifies tangible gaps between impressive specialist results and broader autonomy. It says there are no household robots at scale, no verifiably billion-dollar AI-run company and no visible break in the US unemployment rate. Those observations do not settle whether AI is approaching major economic or scientific change, but they show why a small number of research achievements should not be treated as proof that systems can reliably perform across fields or replace organizations of people.
Products Worth Considering
Philips Sonicare 5950 Rechargeable Electric Toothbrush with 2 Brush Heads, Navy
Blueair Blue Pure 211i Max Air Purifier
Garmin Forerunner 165 GPS Running Smartwatch, 43 mm, Black
As an affiliate, we earn on qualifying purchases.
Two Different ASI Yardsticks
Thompson, an Australian AI consultant who publishes The Memo newsletter, defines artificial general intelligence as a machine performing at the level of an average human. His definition of artificial superintelligence is a machine performing at an expert-human level in practically any field. His tracker ranges from recursive code self-optimization to much more speculative indicators.
That usage differs from other published scales. Google DeepMind’s 2023 “Levels of AGI” framework describes an expert level as performance at the 90th percentile of skilled adults and separates it from “Virtuoso” and “Superhuman” levels. Its June 2026 report describes superintelligence as exceeding the cognitive capabilities of large human organizations. OpenAI’s charter, meanwhile, defines AGI around highly autonomous systems that outperform humans at most economically valuable work. These definitions put different weight on expertise, breadth and autonomy.
Thompson’s tracker, last updated October 26 according to the source, records two achieved, 33 partially achieved and 15 not achieved. In an August correction note, he said “partially” includes documented progress such as funded programmes, research infrastructure, formal scientific missions and enabling work. The review also reports that Thompson called Claude Mythos Preview and GPT-6 Astra “early ASI”; that is his assessment, not an independently established classification.
“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 his tracker
What the Tracker Cannot Establish
The tracker is one researcher’s framework, not a consensus measurement of superintelligence. Its broad definition of ASI does not match the terminology used by every AI lab, and its partial category includes preparatory work as well as progress on the indicators themselves. The review advises treating the list as a set of developments to watch rather than a calibrated measure of how close systems are to superintelligence.
The available source describes the overall verdicts for 18 claims but does not reproduce the complete set of claims or all supporting documentation. It also does not establish whether the results labeled vendor claims have since been independently reproduced. Thompson’s characterization of named models as early ASI remains his view; the review’s source-check does not establish that the systems meet a shared, field-wide standard for superintelligence.
Watch for Independent Replication
The clearest next test for many headline claims is whether independent researchers can reproduce the results and verify how much human guidance was involved. Public proofs, formal certificates and filings can make that assessment easier, but a published result does not by itself establish broad, reliable performance across different fields.
For claims about practical autonomy, the source points to milestones that remain unmet, including reliable multi-step physical tasks and AI systems operating organizations at substantial scale. The review says Thompson separately maintains a conservative AGI countdown at 99%, with its final step awaiting a robot that can assemble something such as flat-pack furniture at average-human reliability. That is a criterion in his framework, not a prediction that the milestone will occur on a particular schedule.
Key Questions
What did the review find about the 18 AI claims?
It classified seven as verified, five as vendor claims, three as based on a single source and three as disputed or contradicted. The source material does not include the complete claim-by-claim table.
How many signs does Thompson’s tracker mark as achieved?
The tracker lists two achieved, 33 partially achieved and 15 not achieved out of 50. Thompson’s definition of partial progress includes funded programmes and other enabling work.
Does the review establish that AI superintelligence has arrived?
No. It reports that Thompson describes certain models as “early ASI,” but that is his assessment. The review stresses that definitions differ and that its findings do not establish a shared, independently verified standard for superintelligence.
Why did some mathematics claims receive a qualified verdict?
The review describes real contributions but also records human feedback, coaching, tool use and researcher involvement in proofs. Those details matter when judging how autonomous a model was.
What developments remain unconfirmed or absent?
The source says household robots have not reached scale, no AI-run company is verifiably worth a billion dollars, and there has been no visible break in the US unemployment rate. It also says some company performance claims have not been independently reproduced in the material reviewed.
Source: ThorstenMeyerAI.com
