Industry & Practical Questions
AI Training Contracts: The Clauses Experts Should Read
A headline rate is only one part of an independent-contractor agreement. You also need to know who owns submitted work, how acceptance is decided, whether…
DecileAI Learn
Clear answers for specialists, ambitious students and experienced contributors deciding where their judgement is worth most.
New to expert AI-training work? Start with the DecileAI Primer.
9 guides found.
Industry & Practical Questions
A headline rate is only one part of an independent-contractor agreement. You also need to know who owns submitted work, how acceptance is decided, whether…
Industry & Practical Questions
A job advertisement can say “remote” and still be closed to somebody working from Lebanon, Kenya or the UAE. Remote usually describes where the work is…
Industry & Practical Questions
Using an AI assistant to complete an AI-evaluation task may sound efficient, but it can defeat the purpose of testing your judgement and may breach the project…
Industry & Practical Questions
The hardest scams to recognise are often built around a believable industry. Genuine AI evaluation work can involve remote interviews and specialised paid…
Industry & Practical Questions
The word “Apply” can hide two very different propositions. One company is trying to staff a funded assignment now. Another is collecting professional profiles…
Industry & Practical Questions
Reinforcement learning from human feedback, or RLHF, describes one family of approaches for improving a model using signals derived from human judgements. A…
Industry & Practical Questions
A gold-standard answer is a carefully checked reference used to establish what good performance looks like on a particular task. It is not necessarily the only…
Industry & Practical Questions
An evaluation rubric is the set of standards contributors use to judge an output. Without one, two experts may both make reasonable decisions but produce…
Industry & Practical Questions
Red teaming is the disciplined attempt to find where a system fails, including failures that normal testing might miss. The work can involve security, misuse…