🔍 Read the full analysis: AI’s Road To AGI And ASI: 50 Signs Of Superintelligence Reviewed 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 prominent claims in researcher Alan D. Thompson’s 50-sign artificial superintelligence tracker: seven are verified, five are vendor claims, three rely on a single source and three are disputed or contradicted. The tracker marks two signs achieved, 33 partially achieved and 15 not achieved, but its definitions and broad use of “partial” do not establish that superintelligence has arrived.
Thorsten Meyer published a review on October 26, 2026, checking 18 claims from AI consultant Alan D. Thompson’s list of 50 signs of artificial superintelligence against research papers, lab statements and other primary sources. Meyer classified seven claims as verified, five as vendor claims, three as supported by a single source, and three as disputed or contradicted; the review does not establish that today’s systems meet a shared definition of superintelligence.
Thompson’s tracker, last updated October 26, marks two indicators achieved, 33 partially achieved and 15 not achieved. The categories cover a wide range, from mathematical results and code improvement to outcomes such as planetary colonization. In an August correction note, Thompson described “partially achieved” as documented material progress, including funded programs, research infrastructure, formal missions and enabling work. That definition means a partial mark does not mean a result is close to completion.
Meyer says the claims with the strongest support tend to involve outcomes that can be checked directly, including mathematical proofs, a physics calculation and regulatory filings. The review describes several mathematicians’ accounts of AI contributions as genuine, while retaining qualifications about human coaching, prior literature and the style of problem. One example is Terence Tao’s account of GPT-5.2’s contribution to an Erdős problem: Tao described it as “more or less autonomously” achieved after feedback, while noting the problem had been misformulated and similar results existed.
Other high-profile claims remain less settled. Meyer labels claims about a set of results generated with GPT-6 Astra as vendor claims, even though Lean certificates are public, because the work has not been independently reproduced in the account. The review also describes recursive AI improvement as real but limited: it cites Anthropic’s report that its best internal model scored 62.8% on a test for which the company said a full replacement for its researchers would need at least 85%. Anthropic estimated AI had made its research and development faster, 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
Why the ASI Labels Matter
The review matters because claims that AI has reached “early ASI” can sound like a settled technical finding when they depend heavily on which definition is being used and how evidence is classified. Thompson defines ASI as a machine performing at expert-human level in practically any field. Meyer notes that this is closer to Google DeepMind’s “Expert” level of AGI than to its later description of superintelligence as exceeding the cognitive capability of large human organisations.
That difference affects how readers interpret the tracker’s status counts and claims about specific models. A model’s contribution to a difficult proof may be meaningful evidence of capability in mathematics, but it does not on its own show broad, reliable performance across fields or autonomous performance at work. The source review also says there is no visible break in the US unemployment rate, no household robots deployed at scale and no verifiable AI-run company worth a billion dollars. Those observations do not settle the question of future capability, but they put boundaries around claims of present-day impact.
For readers, the practical takeaway is to treat the tracker as a watchlist of proposed indicators, not a measurement that independently confirms superintelligence. Its entries mix completed results, company-reported performance, early-stage programs and distant goals. Keeping those categories separate makes it easier to see what has been demonstrated and what remains a prediction or interpretation.
Products Worth Considering
Amazon Kindle Paperwhite (12th Generation, 2024), 16 GB, 7-inch Display, Jade
LG UltraGear 27-Inch QHD OLED Gaming Monitor, 240Hz, 0.03ms, G-SYNC Compatible, FreeSync Premium Pro, HDMI 2.1
Bose SoundLink Flex Bluetooth Speaker (2nd Gen), Portable Waterproof Speaker, Black
As an affiliate, we earn on qualifying purchases.
How the Tracker Defines Progress
Thompson is an Australian AI consultant who publishes The Memo newsletter and maintains a public checklist of 50 indicators, from recursive code self-optimization to speculative milestones. His October video, “ASI: A better world – 2026,” presents a montage of developments from January through October. He has described Claude Mythos Preview and GPT-6 Astra as “early ASI” systems; that is his assessment, not a shared industry determination.
The disagreement begins with the terms. Google DeepMind’s 2023 “Levels of AGI” framework distinguishes performance levels such as Competent, Expert, Virtuoso and Superhuman. Its June 2026 report, “From AGI to ASI,” describes superintelligence as 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 frameworks emphasize different combinations of skill, breadth and autonomy.
Thompson also maintains a separate, conservative countdown to AGI, which the source says stood at 99%. The remaining requirement, in his account, is a robot carrying out multi-step fine assembly, such as assembling flat-pack furniture, at average-human reliability. That is a stated threshold in his framework, not an agreed test for AGI. Meyer’s review focuses on tracing selected claims to their sources rather than independently testing every system or all 50 indicators.
““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
Where the Evidence Still Falls Short
The source review checks 18 selected claims, not every item in the 50-sign tracker, and its classifications do not amount to a systematic benchmark of AI capability. The source material does not identify all details behind every claim in the summary, including the full evidence and independent replication status for each of the 18. A vendor claim may be accurate, but the label means its supporting numbers come from the company rather than an independent reproduction.
There is also no agreed threshold in the cited frameworks for when “early ASI” begins. Thompson’s use of ASI is less demanding than DeepMind’s description of superintelligence, while other definitions emphasize autonomy and economic work. The review does not settle which framework should govern public discussion. Nor does a lack of observed employment change or robots at household scale prove that more advanced systems are absent; these are limits on claims about current, visible effects, not conclusive capability tests.
What Future Checks Can Establish
The next useful evidence will come from independent replication of claimed results, fuller disclosure of methods and clearer distinctions between a model’s unaided work and work completed with human guidance or tools. Public Lean certificates, papers and official filings can help reviewers check specific outputs, but their presence does not by itself establish that a system can reproduce results broadly or autonomously.
Thompson’s tracker is intended to continue following its indicators, and Meyer’s review offers a way to compare selected claims with their underlying sources. Readers should watch for updates to the tracker, new assessments from research labs and external mathematicians, and evidence on the robotics and workplace milestones cited in the review. Until definitions and tests are more consistent, claims that a model has reached ASI should be read with their source, scope and caveats attached.
Key Questions
Did the review conclude that artificial superintelligence has arrived?
No. Meyer checked selected claims and found a mixed evidence record. The review reports that Thompson calls some models “early ASI,” but that is Thompson’s assessment, not an independently established conclusion shared across the cited frameworks.
What does Thompson’s tracker say about its 50 indicators?
As of October 26, it lists two achieved, 33 partially achieved and 15 not achieved. Thompson’s definition of “partially achieved” includes enabling work and funded programs, so those marks do not mean the outcomes are nearly complete.
How many claims did Meyer verify?
Of the 18 selected claims in the review, seven were classified as verified, five as vendor claims, three as resting on a single source, and three as disputed or contradicted.
Why do experts disagree about what counts as ASI?
The cited frameworks set different thresholds. Thompson defines ASI around expert-level performance in practically any field, while Google DeepMind’s 2026 description refers to systems more capable than large human organisations. OpenAI’s charter places emphasis on autonomy and economically valuable work.
What evidence would strengthen claims about AI self-improvement?
Independent replication, transparent methods and results showing that AI can perform the relevant research tasks reliably without extensive human direction would help. In the review, Anthropic reported a 62.8% score on a test it said would require at least 85% for a full replacement for its researchers.
Source: ThorstenMeyerAI.com
