Analyses

Comment acheter des données d'IA qui tiennent la route

24 pieces of operational guidance for teams specifying and buying training data, including the questions worth asking us. Written for the person who signs off the specification, not the one who signs the invoice.

commercial

How to buy AI training data without going through a marketplace

How do I buy AI training data directly rather than through a marketplace?

3 min read

How to price an annotation project without being surprised

How should I budget for an AI data annotation project?

2 min read

compliance

Consent and licensing for field-captured data

How does consent work for AI training data captured in the field?

2 min read

Data residency for AI training data, and why five regions rather than one

How does data residency work for AI training data?

2 min read

The EU AI Act and your training data supply chain

How does the EU AI Act affect AI training data suppliers?

2 min read

Due diligence checklist for AI data suppliers

What should I ask an AI training data vendor before signing?

3 min read

evaluation

Why your speech model fails on real multilingual speech

Why does my speech recognition model fail on multilingual speakers?

2 min read

Building a domain expert panel for medical, legal and financial evaluation

How do I build a panel of domain experts to evaluate my model?

2 min read

Model evaluation, domain experts versus generalists working from a checklist

Should model evaluation be done by domain experts or generalists?

2 min read

Multilingual evaluation, why English-only testing misleads

Why do I need multilingual evaluation for my AI model?

2 min read

Commissioning red teaming, how to get findings you can act on

How do I commission red teaming for my AI model?

2 min read

expert data

Agentic task environments, specifying data for tool-using models

What is agentic task data and how do I commission it?

2 min read

Building a domain expert panel for medical, legal and financial evaluation

How do I build a panel of domain experts to evaluate my model?

2 min read

Expert reasoning traces, what they are and why they cost more

What are expert reasoning traces in AI training data?

3 min read

Model evaluation, domain experts versus generalists working from a checklist

Should model evaluation be done by domain experts or generalists?

2 min read

languages

Sourcing low-resource language data, and why it cannot be shortcut

How do I get training data for a low-resource language?

2 min read

Multilingual evaluation, why English-only testing misleads

Why do I need multilingual evaluation for my AI model?

2 min read

procurement

How to buy AI training data without going through a marketplace

How do I buy AI training data directly rather than through a marketplace?

3 min read

What data provenance actually means, and how to verify it

How do I verify the provenance of AI training data?

3 min read

Data residency for AI training data, and why five regions rather than one

How does data residency work for AI training data?

2 min read

The EU AI Act and your training data supply chain

How does the EU AI Act affect AI training data suppliers?

2 min read

Owned delivery sites versus marketplaces, where the quality difference comes from

What is the difference between an owned delivery site and a data marketplace?

2 min read

How to price an annotation project without being surprised

How should I budget for an AI data annotation project?

2 min read

How a three-tier quality cascade works, and how to check yours

How is quality controlled in a large annotation project?

3 min read

How to run a data pilot before committing real budget

How should I run a pilot with an AI data vendor?

3 min read

Due diligence checklist for AI data suppliers

What should I ask an AI training data vendor before signing?

3 min read

provenance

Consent and licensing for field-captured data

How does consent work for AI training data captured in the field?

2 min read

What data provenance actually means, and how to verify it

How do I verify the provenance of AI training data?

3 min read

EXIF and image provenance, the detail that decides whether a dataset holds up

Why does EXIF metadata matter in AI image datasets?

2 min read

quality

Inter-annotator agreement, what number is actually good enough

What is a good inter-annotator agreement score?

2 min read

How a three-tier quality cascade works, and how to check yours

How is quality controlled in a large annotation project?

3 min read

RLHF preference data, how to specify it so it is worth having

How do I specify RLHF preference data properly?

2 min read

How to write an annotation rubric that survives contact with reality

How do I write a good annotation rubric?

2 min read

robotics

Robotics and embodied AI data, where LiDAR annotation goes wrong

How is LiDAR and robotics training data annotated?

2 min read

specification

Agentic task environments, specifying data for tool-using models

What is agentic task data and how do I commission it?

2 min read

EXIF and image provenance, the detail that decides whether a dataset holds up

Why does EXIF metadata matter in AI image datasets?

2 min read

Expert reasoning traces, what they are and why they cost more

What are expert reasoning traces in AI training data?

3 min read

Inter-annotator agreement, what number is actually good enough

What is a good inter-annotator agreement score?

2 min read

Sourcing low-resource language data, and why it cannot be shortcut

How do I get training data for a low-resource language?

2 min read

Commissioning red teaming, how to get findings you can act on

How do I commission red teaming for my AI model?

2 min read

RLHF preference data, how to specify it so it is worth having

How do I specify RLHF preference data properly?

2 min read

Robotics and embodied AI data, where LiDAR annotation goes wrong

How is LiDAR and robotics training data annotated?

2 min read

How to run a data pilot before committing real budget

How should I run a pilot with an AI data vendor?

3 min read

Speech data collection, what a good specification contains

How do I specify a speech data collection project?

2 min read

How to write an annotation rubric that survives contact with reality

How do I write a good annotation rubric?

2 min read

speech

Why your speech model fails on real multilingual speech

Why does my speech recognition model fail on multilingual speakers?

2 min read

Speech data collection, what a good specification contains

How do I specify a speech data collection project?

2 min read

supply model

Owned delivery sites versus marketplaces, where the quality difference comes from

What is the difference between an owned delivery site and a data marketplace?

2 min read

Un programme à cadrer ?

Apportez le problème plutôt qu'un chiffre de volume. Le volume découle de la spécification.

Voir les capacités