🔍 Read the full analysis: 50 Signs To Assess: How Near Is AI To AGI And ASI? on ThorstenMeyerAI.com
Get hardware and tech essentials 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 of 18 claims in researcher Alan D. Thompson’s 50-indicator ASI tracker found seven verified, five based on vendor claims, three supported by a single source and three disputed or contradicted. Thompson marks two of 50 indicators achieved, but the tracker’s definitions differ from major AI lab benchmarks and many “partial” entries do not mean a system is close to superintelligence.
A review of 18 prominent claims in researcher Alan D. Thompson’s 50-sign tracker found evidence of meaningful AI contributions in mathematics and other fields, but no clear basis for treating today’s leading models as established artificial superintelligence. The tracker lists two indicators achieved, 33 partially achieved and 15 not achieved; the review says seven of the claims it checked were verified, while others relied on company assertions, single sources or disputed evidence.
Thompson, an Australian AI consultant who publishes The Memo newsletter, maintains a public checklist of signs that artificial superintelligence may have arrived. Its entries range from recursive code self-optimization to much more speculative milestones. In his October video, “ASI: A better world – 2026,” he presents developments from January through October and describes Claude Mythos Preview and GPT-6 Astra as “early ASI.” Those descriptions are Thompson’s assessment, not a consensus finding or a standardized measurement.
The review tested 18 of the montage’s strongest claims against papers, researchers’ accounts, lab posts and other primary material. It classified seven as verified, five as vendor claims, three as single-source claims, and three as disputed or contradicted. Verified examples included mathematical work discussed by Terence Tao and Don Knuth, and a conjecture disproof attributed to an OpenAI model and examined by external mathematicians. Some of these accounts include important limits: human feedback or coaching was involved, and the work’s scope or originality needs qualification.
Other headline results remain tied to company reporting. For example, the review labels OpenAI’s account of GPT-6 Astra producing ten results on decade-old problems for about $2,000 a vendor claim, while noting that Lean certificates were made public and that humans prepared the problems. The material supplied for this report does not provide enough detail to assess every item in the 50-sign checklist independently.
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
What the Checklist Can Measure
The tracker brings attention to developments that may matter, including AI-assisted mathematics, research automation and systems that can improve software. But its status labels should not be mistaken for a calibrated estimate of how near AI is to AGI or ASI. Thompson says “partially achieved” includes funded programmes, research infrastructure, scientific missions and enabling work. That category can signal activity toward an indicator without showing that a model has reached the capability itself.
The review’s findings also show why checking claims against original evidence matters. A model’s contribution to a proof may be substantial while still involving researcher guidance or a problem that was initially misformulated. Company-reported benchmark results can be informative, but they are not equivalent to independent replication. For readers following AI progress, the useful distinction is between demonstrated capability, a lab’s own report and a broader interpretation about what that capability means.
Claims about superintelligence carry consequences beyond terminology. They can shape public expectations, policy discussions and decisions about AI deployment. A tracker can serve as a watchlist, but a label such as “early ASI” should be read alongside the author’s definition, the quality of evidence and what the system can do reliably outside a narrow test or research task.
Products Worth Considering
Logitech MX Mechanical Mini Mac Wireless Keyboard Low-Profile – Space Grey
Apple MacBook Pro 14.2-inch with M5 Max, 18-Core CPU, 20-Core GPU, 48GB Memory, 2TB SSD, Space Black
Logitech MX Keypad Programmable Macro Pad, Graphite
As an affiliate, we earn on qualifying purchases.
Three Different AGI Yardsticks
Part of the disagreement is definitional. Thompson defines AGI as a machine performing at the level of an average human and ASI as one performing at the level of an expert human in practically any field. Under that definition, a system that produces expert-level work in a domain may count as ASI even if its abilities are uneven elsewhere.
Google DeepMind’s 2023 “Levels of AGI” framework uses categories that include “Competent,” “Expert,” “Virtuoso” and “Superhuman.” In that scale, expert performance is one level below superhuman. 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 frameworks therefore ask different questions about capability, breadth and autonomy.
Thompson also maintains a separate “conservative countdown to AGI” that he puts at 99%. He says the remaining step is a robot able to complete multi-step fine assembly, such as building IKEA furniture, with average-human reliability. That figure is his estimate, not an agreed industry measure. It should not be confused with the 50-indicator tracker or with a laboratory test that establishes AGI.
“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
Evidence Still Has Major Gaps
The review’s 18 claims are a selected sample, not an independent audit of every entry in the tracker. The supplied source does not include the full evidence table or enough information to verify all findings described in the checklist. In particular, vendor claims have not necessarily been reproduced independently, and a single-source account may not establish how reliably a capability works in other settings.
It is also unclear how consistently the tracker applies its labels across different kinds of indicators. A mathematical result, a funded scientific mission and the arrival of household robots are not comparable milestones, yet they appear within the same 50-item framework. The source says there are no household robots at scale, no verifiably billion-dollar AI-run company and no visible break in the U.S. unemployment rate, but it does not supply measurements or define thresholds for those statements.
There is no shared, accepted threshold for AGI or ASI in the material reviewed. The competing definitions mean that a conclusion can change depending on whether the test is expert performance in a field, broad economic capability, autonomy or superiority to large human organisations. Thompson’s “early ASI” label remains his interpretation, not a settled classification.
What Future Evidence Should Show
The next useful developments are independent checks of high-profile model results, clearer reporting of how much human direction a task required, and repeated evaluations that test whether performance generalizes beyond a demonstration. Public proofs and formal verification can help in mathematics, while other claims may need transparent benchmarks, replication or operational evidence.
Thompson’s tracker is listed as updated on October 26, 2026, but the supplied material gives no date for a forthcoming update or independent review. Readers can watch for changes to the achieved and partial counts, while checking whether the underlying evidence has changed—not only the status label. The central question remains whether systems can perform a wide range of tasks reliably and autonomously, rather than whether isolated signs of progress can be assembled into a single milestone.
Key Questions
What does the 50-sign ASI tracker report?
Thompson’s tracker lists two indicators achieved, 33 partially achieved and 15 not achieved as of October 26, 2026. The categories cover a range of proposed signs, and “partially achieved” can include enabling work rather than a completed capability.
Does the review establish that current AI is ASI?
No. The review reports several verified AI contributions but also identifies claims based on vendor reporting, a single source, or disputed evidence. “Early ASI” is Thompson’s characterization, and the answer depends in part on which definition of ASI is used.
How many claims did the review verify?
Of the 18 selected claims, the review classified seven as verified, five as vendor claims, three as single-source claims and three as disputed or contradicted. Those figures describe the selected claims, not the full 50-item tracker.
Why do definitions of AGI and ASI matter?
Thompson defines ASI as expert-human performance in practically any field, while other frameworks set different standards involving superhuman performance, autonomy or capability beyond large human organisations. A system can meet one yardstick without meeting another.
What evidence would make future claims stronger?
Useful evidence would include independent replication, clear records of human involvement, transparent evaluation methods and demonstrations that capabilities work reliably across tasks and settings. A company’s own benchmark result may be informative, but it is not the same as an independently reproduced result.
Source: ThorstenMeyerAI.com
