🔍 Read the full analysis: 50 Signs Of Superintelligence: Is AI Approaching AGI Or ASI? on ThorstenMeyerAI.com
Prime made for students and young adults
- Fast, free delivery for dorm and study essentials
- Prime Video and Amazon Music included
- Member-only deals
TL;DR
Researcher Alan D. Thompson’s October 26 tracker marks two of 50 signs of artificial superintelligence achieved, 33 partially achieved and 15 not achieved. A review of 18 prominent claims finds seven verified, but the tracker’s definition of superintelligence is far less demanding than some major AI labs’ definitions.
A review of researcher Alan D. Thompson’s 50-sign superintelligence tracker found a mixed record behind claims that leading AI models are already “early ASI”: of 18 prominent claims examined, seven were verified, five depended on company-reported results, three relied on a single source and three were disputed or contradicted. Thompson’s tracker, updated October 26, lists two indicators as achieved, 33 as partially achieved and 15 as not achieved.
The tracker ranges from mathematical breakthroughs to ambitious milestones such as household robots, AI-run companies and even faster-than-light travel. Thompson says “partially achieved” includes funded programmes, research infrastructure and other enabling work. That status does not mean a milestone is close to completion: it can reflect preparatory activity rather than the capability itself.
The review found its strongest support in claims that can be checked against papers, expert accounts or official records. Mathematicians Terence Tao, Don Knuth and Tim Gowers have described real AI contributions, though the examples included coaching, human-written proofs, existing literature or limitations tied to the problem. Other claims, including a set of results attributed to an OpenAI model, were classified as vendor claims because public certificates were available but the underlying work had involved human preparation.
The most significant capability gap in the source’s account concerns AI improving AI research. Anthropic’s own internal evaluation put its best model at 62.8% on a test the company says a full replacement for its researchers would require at least 85% on. Anthropic also estimated that AI had made its research and development faster, but “not yet by a factor of 2.” These are company-reported assessments, not independent measurements.
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 Status Labels Matter
The tracker is a useful catalogue of developments to watch, but its headline percentages cannot be read as a measured distance to superintelligence. The meaning of “partially achieved” is broad, and Thompson’s definition of ASI differs substantially from definitions used by major AI labs. Readers could otherwise mistake evidence of research activity for evidence that a system already has the capability being tracked.
The underlying question matters beyond terminology. If AI systems can reliably contribute to research, software development or other expert work, they could affect scientific progress and the labour market. But the review points to a gap between demonstrations in selected tasks and dependable, autonomous performance across many domains. It also reports no household robots operating at scale, no verifiably billion-dollar company run by AI, and no visible break in the US unemployment rate tied to AI. Those observations do not settle AI’s future economic effects; they show that broad real-world consequences should not be inferred from isolated examples.
Popular right now
Breville Bambino Espresso Machine, Brushed Stainless Steel
De’Longhi Classic Matte Black Espresso Machine with Milk Frother
Ninja Flip Toaster Oven and Air Fryer, 8-in-1, Stainless Steel, SP151
As an affiliate, we earn on qualifying purchases.
Two Different Superintelligence Scales
Thompson defines AGI as a machine performing at the level of an average human and ASI as one performing at an expert-human level in practically any field. That makes his ASI threshold closer to Google DeepMind’s “Expert” level of AGI than to the highest levels in DeepMind’s 2023 framework, which also describes “Virtuoso” and “Superhuman” performance.
DeepMind’s June 2026 report uses a more demanding description of superintelligence: 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. The differences put autonomy and breadth at the centre of the debate: a model that helps solve a difficult problem may be impressive without meeting a standard based on independent performance across fields.
Thompson’s tracker is an individual researcher’s framework, not an agreed scientific measurement. His separate “conservative countdown to AGI” stood at 99% in the source material, with its final step tied to a robot completing multi-step fine assembly at average-human reliability. That estimate reflects his own chosen milestone and should not be treated as a consensus forecast.
““documented material progress toward an indicator, including funded programs, research infrastructure, formal scientific missions, and enabling work””
— Alan D. Thompson, describing the tracker’s “partially achieved” category
What the Tracker Cannot Establish
The review covers 18 selected claims, not all 50 indicators, and its findings do not establish a single agreed threshold for AGI or ASI. The source does not provide enough detail here to independently assess every claim or reproduce the review’s verification process. “Verified” also applies to particular reported results; it does not show that a model has general expertise or autonomous capability across fields.
Company figures remain claims from the companies unless independently reproduced, while single-source findings and disputed entries carry their own limitations. It is also unclear how consistently the tracker applies its categories across very different indicators. The claim that leading models are “early ASI” is Thompson’s assessment, not a consensus finding, and depends on using his comparatively lower threshold for ASI.
Evidence to Watch in New Tests
The next useful developments would be independent reproductions of company-reported results, clearer reporting on how much human guidance a model received, and tests of whether performance carries over to unfamiliar tasks. For claims about AI helping AI research, researchers and companies would need to publish methods and results that allow outside observers to assess the scale of any productivity gains.
Readers can also watch for concrete milestones that go beyond research preparation: reliable multi-step work by robots in ordinary settings, sustained autonomous performance across domains, and measurable effects in workplaces. Until such evidence is available, the tracker is best read as one researcher’s map of possible indicators—not proof that superintelligence has arrived.
Key Questions
How many indicators does Thompson say have been achieved?
The tracker lists two achieved, 33 partially achieved and 15 not achieved, as of October 26, 2026. Thompson’s partial category includes enabling work and funded programmes, not just completed capabilities.
Does the review confirm that AI has reached superintelligence?
No. It reports that seven of 18 selected claims were verified, but those results do not establish broad, autonomous expert performance. The “early ASI” label is Thompson’s view and depends on his definition.
Why do definitions of ASI differ?
Thompson defines it around expert-level ability in practically any field. Google DeepMind’s 2026 description sets a higher bar, comparing a system’s capabilities with those of large organisations of humans. OpenAI’s definition of AGI also stresses autonomy and economically valuable work.
What evidence did the review find for AI contributions to mathematics?
The source review cites accounts from mathematicians including Terence Tao, Don Knuth and Tim Gowers. It describes genuine contributions, while noting caveats such as human coaching, human-written proofs and the role of existing mathematical literature.
What remains a major test of AI capability?
The source identifies dependable, multi-step fine assembly by a robot as a milestone in Thompson’s separate AGI countdown. More broadly, independent tests of autonomy, reliability and performance across unfamiliar tasks remain important.
Source: ThorstenMeyerAI.com
