Metatron Intelligence
Today: the five AI headlines of the day → and the AI Wiki
What an entity record is How to read this page

A named actor, followed across desks

An organisation, model, chip or programme, with the statements attached to it, so that the same actor can be tracked without re-reading every domain.

Statements carry their own sources

Each statement on this page is bound to a source and a date. The entity page collects them; it does not average them into a verdict.

Absence is not evidence

An entity with few statements has been observed less, not necessarily done less. Read the count as coverage, never as significance.

New to this publication?

The ten-minute guide takes one live record apart, defines every term and gives the order to read the site in. Start here →

ENT-9a3c06c8-568b-4f39-9f62-e1edb55e8a6a · benchmark · graduate science multiple choice benchmark

GPQA

Reviewed identifier idavidrein/gpqa@56686c06f5e19865c153de0fdb11be3890014df7 · current status intentionally unknown

Scope

Reviewed benchmark or evaluation-protocol identity; model results and leaderboard ranks are separate, unrepresented observations.

Boundary

The `Google-Proof` title is the benchmark name, not a guarantee that every item is unsearchable, uncontaminated or representative of all expert reasoning.

Benchmark comparability passport

Publisher
GPQA authors
Version
source revision 56686c06; no semantic benchmark version stated
Task scope
Graduate-level, Google-proof question-answering dataset with separate data files and baseline implementations.
Metric contract
A result is meaningful only with the exact data file/subset, shuffled answer-choice seed and prompt/retrieval mode; no single aggregate is admitted here.
Evaluation conditions
Result identity includes repository/data revision, subset, model, zero/few-shot or chain-of-thought mode, closed/open-book retrieval setting, answer-choice shuffle seed and cache behavior.
Comparability boundary
Do not combine GPQA subsets or compare retrieval and closed-book runs, different prompts, seeds or code revisions as one measurement.
Contamination boundary
The dataset includes a canary string; this is a detection/deterrence aid, not proof that training exposure did not occur.

This passport describes a benchmark definition or protocol. It publishes no model score, rank or cross-version equivalence.

Source-qualified relations

  1. No reviewed relation is asserted for this identity.

Atlas claims with an exact source route

  1. No exact source-URL connection in r0056. No name match was substituted.

Every displayed attribute links to source-qualified statement IDs. Relations are explicit and source-qualified. Identity does not imply ownership, legal status, present availability or equivalence with a mutable alias.