A model cannot tell a checked figure from an unchecked one
Calaa is the group's intelligence layer. What decides its quality is not which model it uses, which is the part that distinguishes no one, but which corpus sits underneath. Mind checks that corpus before it goes in.
Por qué la comprobación va antes del modelo
Hoy cualquiera genera un informe con una inteligencia artificial en diez minutos. Lo que ya no es fácil es responder de dónde sale cada dato, y ese es el trabajo que hace Mind.
Un modelo entrenado sobre cifras sin origen produce respuestas igual de fluidas que uno entrenado sobre cifras comprobadas. La diferencia no se nota al leerlas. Se nota seis meses después, cuando alguien decide con una de ellas.
The two answers
Both read the same. Only one can be checked
- the source
- the date
- the parameter
Every figure keeps the three marks
Where does this figure come from?
Answered in one click
- the source
- the date
- the parameter
The answer reads just as fluently and there is nothing underneath
Where does this figure come from?
Cannot be answered
The difference does not show when you read them. It shows when someone asks where a figure comes from, and by then the decision has been taken.
Two answers to the same brief, one on a checked corpus and one on an unchecked one. What changes is not in the text, it is in what each figure carries underneath. The figure compares structure, not writing.
Las tres reglas que se aplican antes de escribir una línea
Se aplican a cada dato que va a entrar en el corpus, no al informe terminado.
- 01
Fuentes fiables
Instituciones oficiales, organismos públicos, estudios académicos, publicaciones especializadas y empresas reconocidas de investigación. Cuando la única fuente disponible no cumple ese listón, el dato entra marcado o no entra.
- 02
Contraste cruzado
Cada dato se cruza con varias fuentes para descartar el sesgo. Cuando dos fuentes fiables no coinciden, se conservan las dos en vez de elegir la que conviene.
- 03
Análisis crítico
Se evalúa la fuente, la fecha, el método y los intereses que pueda haber detrás. Un dato correcto en el contexto equivocado también engaña, y es la forma más frecuente de equivocarse con un número que es cierto.
The three gates
None of the three rules answers yes or no
A figure
- 01
Fuentes fiables
Enters marked, or does not enter
- 02
Contraste cruzado
If two reliable sources disagree, both are kept
- 03
Análisis crítico
A correct figure in the wrong context stays out
Enters the corpus, with its label
All three come before writing, not after. A figure that reaches the end enters carrying the label of where it passed, and that label travels with it into the corpus.
A figure's route through the three rules, in the order they are applied. Each gate has its own exit. It ranks the route, not how many figures each one takes.
01.1Before the model
Un modelo no pregunta de dónde sale un dato. Alguien tiene que haberlo preguntado antes.
Checking has someone behind it. Someone decides what goes in, and that decision is written next to the figure.
Qué se marca, y por qué el modelo lo necesita marcado
Tres etiquetas que viajan con el dato y no se pierden al agregarlo.
This is an estimate
An estimate presented as a measurement spreads without warning.
When the figure is a calculation of ours and not a measurement, it says so and explains how it was obtained. After that it cannot be corrected, because whoever received it used it as if it were a taken reading.
This cannot be answered
No artificial intelligence writes on its own that something has no answer.
A question the available data does not close is marked as open, with what it would take to close it. A model always finds something to answer, so the gap has to be written down by someone first.
This comes from here
The mark travels with the figure and does not stay in an appendix.
Every figure keeps its source, its date and the parameter it was accepted by, next to the figure. That way it is still there when someone takes it elsewhere, which is when it usually gets lost.
Lo que Calaa devuelve
La capa de inteligencia agrega, relaciona y responde sobre un corpus que Mind ha comprobado. Cuando una respuesta se apoya en un dato marcado como estimación, la marca sigue ahí. Esa es la única manera de que el grupo pueda usar su propia inteligencia para decidir y no solo para redactar.
The route of the mark
The mark is still there when the answer comes out the other side
The mark
The checked figure
Enters with its source, its date and the parameter it was accepted by
The corpus
Lost here if it does not travel declaredIt is aggregated with everything else, and that is where the origin usually gets lost
The model
Learns on what has been checked, not on what was at hand
The answer
Arrives with the mark of what supports it, estimates included
Without the mark, the answer reads just as well and cannot be checked. With it, the group decides with its own intelligence instead of only writing with it.
The four hops a figure takes from being checked to supporting an answer. The figure follows one mark; it does not count how many figures pass through each layer.
