🔍 Read the full analysis: How Close Is AI To Superintelligence? 50 Signs Point Toward AGI And ASI on ThorstenMeyerAI.com
Get business pricing on tech for your team
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
TL;DR
An October 26 review of researcher Alan D. Thompson’s 50-item ASI tracker reports two indicators marked achieved, 33 partially achieved and 15 not achieved. The review says some AI mathematics results have independent support, while several high-profile capability and self-improvement claims rely on company statements or limited evidence. The tracker’s “early ASI” label uses a definition far less demanding than some major labs’ definitions of superintelligence.
A review published October 26 by Thorsten Meyer examined 18 claims in AI consultant Alan D. Thompson’s list of 50 signs of artificial superintelligence, finding seven verified, five based on vendor claims, three supported by a single source and three disputed or contradicted. Thompson’s tracker marks two indicators achieved, but the review says its broad definitions and partial-progress category do not establish that superintelligence has arrived.
Thompson, an Australian AI consultant who publishes The Memo newsletter, maintains a public checklist covering developments from mathematical proofs to robotics and space exploration. Its status grid, last updated October 26, lists 33 indicators as partially achieved and 15 as not achieved, alongside the two achieved. In an August correction note, Thompson defined partial progress broadly to include funded programmes, research infrastructure, formal scientific missions and other enabling work. That designation does not mean a capability is nearly complete.
Meyer’s review checked selected claims against papers, researcher comments, company posts and official filings, and assigned five kinds of verdict: verified, vendor claim, single source, disputed or contradicted, and not found. The review describes AI contributions to mathematics as among the better-supported examples. It cites assessments by mathematicians Terence Tao, Don Knuth and Tim Gowers, while retaining qualifications about human guidance, existing literature and the types of problems involved.
Other claims receive more cautious treatment. The review says a reported GPT-6 Astra set of mathematical results was a vendor claim, even though Lean certificates were public, because the people and process behind the results still matter to assessing independence. For recursive improvement, it cites Anthropic’s internal assessment: the company’s best model scored 62.8% on a test, compared with at least 85% that Anthropic said would be needed to replace its researchers fully. Anthropic also estimated that AI 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 Definitions Change the Claim
The central issue is not just whether models can solve difficult problems, but what level of capability counts as superintelligence. Thompson defines artificial general intelligence as performing at the level of an average human and artificial superintelligence as performing at expert-human level in practically any field. Under that definition, strong performance in expert tasks can support an “early ASI” label.
Other frameworks set a higher bar. Google DeepMind’s 2023 scale distinguishes competent, expert, virtuoso and superhuman performance. Its 2026 report describes superintelligence as more capable than large organisations of humans. OpenAI’s charter, meanwhile, defines AGI around highly autonomous systems that outperform people at most economically valuable work. These differences make headline labels difficult to compare: expert-level work in some fields is not the same as broad, autonomous superiority.
For readers, that distinction affects how to interpret claims about jobs, scientific discovery and the pace of AI development. A checklist can help track research, but its coloured categories are not a standardized scientific measurement. The review also notes that household robots are not operating at scale, that it found no verifiable AI-run company worth a billion dollars, and that there is no visible break in the US unemployment rate attributable to AI. Those observations do not settle AI’s future effects, but they caution against treating capability demonstrations as proof of economy-wide transformation.
Products Worth Considering
Sony WH-1000XM4C Wireless Noise-Cancelling Headphones, Black
DJI Osmo Mobile 7P with Mic Mini TX (Infinity Black) Gimbal Stabilizer
LG 34GX90SA-W 34-inch Ultragear WQHD OLED Curved Gaming Monitor
As an affiliate, we earn on qualifying purchases.
How the ASI Tracker Scores Progress
Thompson’s checklist ranges from recursive code self-optimization to far-reaching outcomes such as travel beyond Earth. His October video, “ASI: A better world – 2026,” presents a montage of entries from January through October. He has described Claude Mythos Preview and GPT-6 Astra as “early ASI” systems, using his own definition rather than a common industry standard.
The tracker’s broad partial-progress category can include work that enables a future indicator rather than achieving it. Meyer points to two exoplanet candidates reportedly found with Claude Code that were placed under an indicator concerning the first planet beyond Earth to be colonized. That example illustrates why a partial status should not be read as evidence that the listed outcome has occurred.
Thompson also publishes a separate countdown to AGI that he puts at 99%. He says the remaining step is a robot able to perform multi-step fine assembly, such as assembling flat-pack furniture, with average-human reliability. This is a separate estimate, not an independently validated measure of progress or a consensus forecast.
“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 Evidence Cannot Establish
The source review is a check of 18 selected claims, not a comprehensive audit of all 50 indicators or a formal measurement of AI capability. Its findings depend on the evidence available for each claim and on its chosen categories. A result labeled verified may confirm a specific mathematical contribution; it does not by itself establish general intelligence or superintelligence.
Some company-reported results have not been independently reproduced, while other claims rest on a single source or are contested. The tracker’s “early ASI” description is Thompson’s judgment, and definitions used by researchers and laboratories differ. It is also unclear how quickly current systems will improve, whether benchmark performance will translate into reliable real-world autonomy, or when robots will handle complex work at human-level reliability.
Evidence to Watch in Future Updates
Further independent checks of mathematical and scientific results could clarify which reported advances are reproducible and how much human guidance they required. For claims based on company testing, publication of methods, evaluation details and independent replication would make comparisons more informative. Progress on reliable multi-step robotics is another concrete milestone cited by Thompson’s own countdown.
Until then, the tracker’s next updates may document additional research or enabling programmes, but those entries should be distinguished from completed capabilities. Readers assessing future claims can compare the exact task, evidence and definition being used rather than relying on an “AGI” or “ASI” label alone.
Key Questions
Does the review conclude that AI superintelligence has arrived?
No. It reports that Thompson’s tracker marks two of 50 indicators achieved, while emphasizing that the tracker is one researcher’s framework and does not establish that AI meets other definitions of superintelligence.
What does “partially achieved” mean in Thompson’s tracker?
Thompson says it includes documented progress such as funded programmes, research infrastructure, scientific missions and enabling work. It does not necessarily mean the capability is close to completion.
How did the review assess the claims?
Meyer checked 18 selected claims against primary sources, including papers, researcher statements, company posts and official filings. He classified them as verified, vendor claims, single-source claims, disputed or contradicted, or not found.
Why do researchers disagree about whether AI is already ASI?
They use different definitions. Thompson’s ASI threshold is expert-human performance in practically any field, while Google DeepMind’s 2026 description compares a system with the capabilities of large human organisations. The same model can receive different labels under those standards.
What evidence would help clarify the claims?
Independent replication, transparent evaluation methods and demonstrated reliability on real-world tasks would strengthen claims about AI capability. For robotics, a clear milestone would be reliable performance on complex, multi-step physical work.
Source: ThorstenMeyerAI.com
