🔍 Read the full analysis: 50 Signs Of Superintelligence, Explained: AI’s Distance From AGI And ASI on ThorstenMeyerAI.com
Get everyday 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-sign superintelligence tracker found seven verified, five based on vendor claims, three supported by a single source and three disputed or contradicted. The tracker lists two of 50 indicators as achieved, but its broad definition of “partial” progress and its use of “ASI” differ from other frameworks.
A review published October 26, 2026, checked 18 claims in AI consultant Alan D. Thompson’s public tracker of 50 signs of artificial superintelligence, finding a mix of independently supported results, company-reported achievements and disputed claims. Thompson’s tracker marks two indicators achieved, 33 partially achieved and 15 not achieved, but its definitions and status rules do not match every major AI lab’s measures of AGI or superintelligence.
The review, published by ThorstenMeyerAI.com, traced selected claims from Thompson’s October video montage and tracker to papers, researcher accounts, company posts and regulatory filings. It classified seven of 18 claims as verified, five as vendor claims, three as relying on a single source, and three as disputed or contradicted. The article says the strongest evidence tended to involve results that could be checked, including mathematics, a physics calculation and formal records. These findings apply to the 18 claims reviewed, not all 50 indicators.
Several mathematical claims had support from researchers who described both the AI contribution and its limits. Terence Tao reported that GPT-5.2 Pro, used with the tool Aristotle, solved an Erdős problem “more or less autonomously” after feedback; the problem had been misformulated, and similar results existed. Donald Knuth wrote that Claude found a construction for odd sizes during coached exploration, while Knuth wrote the proof. A separate claim that an OpenAI model disproved a 1946 Erdős unit-distance conjecture was supported by a paper involving external mathematicians, according to the review.
Other entries were less conclusive. The review classed OpenAI’s reported results for GPT-6 Astra as a vendor claim: Lean certificates were public, but the results had not been independently reproduced in the account. The article also highlights Anthropic’s assessment of AI-assisted research: its best internal model scored 62.8% on a test for which the company said a full replacement for its researchers would require at least 85%. Anthropic said 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
Evidence Behind the ASI Claims
The tracker brings attention to real AI progress, especially in mathematical research, but its labels can suggest more certainty or proximity to superintelligence than the underlying evidence supports. A claim marked partially achieved may mean that a funded program, research infrastructure or enabling work exists, not that the stated capability is close to delivery. Thompson’s August correction note defines the category broadly, making the distinction between progress and achievement important for readers.
Definitions also affect how the results are interpreted. Thompson describes ASI as expert-human performance in practically any field. Google DeepMind’s 2023 framework labels expert-level performance as “Expert AGI”, while its June 2026 report describes superintelligence as capability beyond large organisations of humans. OpenAI’s Charter instead defines AGI around highly autonomous systems that outperform people at most economically valuable work. A system might meet one description in a particular domain without meeting another’s broader standard.
That matters because declarations that AI is already “early ASI” are claims, not a settled scientific classification. The reviewed evidence supports meaningful advances in selected tasks; it does not establish that current systems have broad, autonomous expert capability across fields or exceed the combined capacity of large human organisations.
Products Worth Considering
Logitech MX Mechanical Wireless Illuminated Keyboard, Tactile Quiet, Graphite
LG UltraGear 27-Inch QHD OLED Gaming Monitor, 240Hz, 0.03ms, G-SYNC Compatible, FreeSync Premium Pro, HDMI 2.1
DJI Osmo Mobile 8 Standard Combo Phone Gimbal Stabilizer
As an affiliate, we earn on qualifying purchases.
How the Tracker Defines Progress
Thompson, an Australian AI consultant who publishes The Memo newsletter, maintains a public checklist ranging from recursive code improvement to claims far beyond current capabilities. Its October 26 status grid records two achieved indicators, 33 partial and 15 not achieved. The tracker’s examples and categories should be read as Thompson’s framework, rather than an agreed industry benchmark.
The article points to a concrete example of how expansive “partial” can be: two exoplanet candidates identified with Claude Code were placed under an indicator about the first planet beyond Earth being colonized. That classification records what Thompson regards as enabling progress; it does not mean that a planet has been colonized. The tracker also calls Claude Mythos Preview and GPT-6 Astra “early ASI,” a characterization attributed to Thompson, not independently established by the review.
Thompson separately estimates that his conservative countdown to AGI stands at 99%, with the final percentage tied to a robot completing multi-step fine assembly, such as building IKEA furniture, at average-human reliability. This is his estimate and threshold, not a measured consensus. The review says that household robots are not operating at scale, that no AI-run company has been verifiably shown to be worth $1 billion, and that no visible break in the U.S. unemployment rate has been established as evidence of AI-driven change.
“More or less autonomously”
— Terence Tao, as quoted in the review
What the Evidence Cannot Establish
The review does not settle how close AI is to AGI or ASI. Its audit covers 18 selected claims, and the supplied account does not provide a full independent assessment of every tracker entry. Some claims depend on company-reported results, while others have limited sourcing or are disputed. A public certificate or an expert’s account can support a particular result without showing that a system can perform similar work reliably across unrelated tasks.
It also remains unclear whether any current model meets a shared definition of AGI, since the definitions cited place different weight on human-level performance, expert ability, autonomy and breadth. The review reports no household robots operating at scale, no verifiably billion-dollar AI-run company and no visible break in the U.S. unemployment rate. Those observations do not by themselves prove that AI has had no economic or practical effects; they indicate that the specific milestones in the article have not been established.
Tests That Could Change the Picture
The next useful evidence will be independent replication of results that currently rest on company claims, alongside more transparent reporting on how much human coaching, tool use and verification each result required. For the AGI countdown Thompson describes, a notable milestone would be a robot reliably completing multi-step fine assembly at average-human performance. The source does not give a date for that milestone.
Thompson’s tracker is set up to change as he adds evidence, but its status grid remains one researcher’s assessment. Readers can watch for updates to the tracker and for new papers, independent evaluations and lab reports. Until definitions and evaluation methods are more consistent, claims that AI has reached AGI or ASI need to be judged against the specific capability and evidence being presented.
Key Questions
How many signs does Thompson’s tracker say have been achieved?
The tracker’s status grid, dated October 26, 2026, lists two indicators as achieved, 33 as partially achieved and 15 as not achieved.
Does “partially achieved” mean a system is nearly there?
No. Thompson says the category can include funded programs, research infrastructure, scientific missions and other enabling work. It does not necessarily mean the capability itself is close to completion.
Did the review find that current AI is superintelligent?
No. It reports selected advances and attributes the “early ASI” description to Thompson. The review says that definitions differ and the evidence does not establish that current systems meet a broad, shared standard for superintelligence.
What was the review’s assessment of AI mathematics claims?
It found support for several contributions, including accounts from Terence Tao and Donald Knuth, while also describing human feedback, coaching or proof-writing involved in some cases. The review treats the claims individually rather than as proof of general mathematical autonomy.
Source: ThorstenMeyerAI.com
