EXHIBIT 09 / CLAIM READING

When someone says “AGI,” what exactly are they claiming?

AGI is not a product category with one accepted pass line. A useful claim has to say which tasks count, how performance is compared, and whether it is describing capability or a chosen level of autonomy.

SHORT ANSWERAsk for the definition, the task set, and the evidence.
5–8 MIN · MANUAL CONTROLS · NO ACCOUNT

ROOM 01 / DEFINE

One headline often collapses three independent questions.

  1. 01 / BREADTH How much of the task world?

    Generality concerns the range of tasks on which a system reaches a stated performance threshold—not the number of features in one product.

  2. 02 / DEPTH How well on those tasks?

    Performance must be defined against a comparison group, metric, and threshold. One spectacular score cannot stand in for performance across most tasks.

  3. 03 / AUTONOMY Who controls the work?

    A capable model can be deployed as a tool, consultant, collaborator, expert, or agent. The interface and permissions choose autonomy; capability does not force it.

  4. 04 / EVIDENCE What was actually measured?

    A benchmark result supports only its tested tasks, conditions, and scoring rules. Product popularity, fluency, or confidence is not an AGI measurement.

ROOM 02 / INSPECT

Turn “AGI arrived” into a claim that could be checked.

Advance the four-part claim audit. It is an editorial method derived from the paper’s dimensions, not a benchmark result.

MANUAL EVIDENCE PATHNOT A CAPTURED SYSTEM TRACE

TERM

Name the definition

Does “AGI” mean broad human-level performance, economically valuable work, learning new skills, or something else?

WHAT THIS LETS YOU SAY

Without a definition, two people can use the same word for different thresholds.

Step 1 / 4

Keyboard: focus this instrument and use ← or →. It never advances by itself.

ROOM 03 / TEST

Place a hypothetical claim inside one published framework.

Change three labels. The browser names the corresponding cell and interaction style; it does not test a real system or decide whether AGI exists.

LIVE BROWSER CLASSIFIERNO MODEL, HARDWARE, ACCOUNT, OR REMOTE SERVICE

SCENARIOHypothetical evidence card—not a score for any named model.

01Capability breadth

A clearly scoped task or family of tasks.

02Minimum performance over most selected tasks

Roughly at or above an unskilled human comparison in the paper’s terminology.

03Chosen interaction style

The human controls the task and invokes bounded automation.

FRAMEWORK READING

This remains a Narrow AI cell.

Changing performance or autonomy does not make narrow evidence general. The task coverage would need its own support.

SELECTED FACTS
  • A clearly scoped task or family of tasks.
  • Roughly at or above an unskilled human comparison in the paper’s terminology.
  • The human controls the task and invokes bounded automation.

The authors’ table contains more levels and qualifications than this teaching control. Read the source before applying the labels.

ROOM 04 / VERIFY

Know where the framework ends and the museum begins.

PUBLISHED FRAMEWORK

Breadth × performance

The paper proposes Narrow/General breadth and five performance levels, and says unambiguous classification needs standardized, ecologically valid benchmarks.

PUBLISHED DISTINCTION

Capability is not autonomy

The paper treats autonomy as an interaction and deployment choice that may be enabled—but is not determined—by capability.

NOT CLAIMED

A winner or arrival date

The museum does not declare that any current model reached AGI, or that this ontology is the accepted definition.

REVIEWED SOURCES

Open the record before repeating the claim.

Levels of AGI for Operationalizing Progress on the Path to AGI arXiv (Google DeepMind) · key-sections · checked 2026-09-11 Abstract, the six principles in section 3, the breadth-by-performance levels in section 4, and capability versus autonomy in section 6; every citation was not rechecked and the authors’ framework is not treated as an industry standard.
NEXT / FROM CLAIMS TO COMPUTATIONSee where a model becomes concrete operations →