IJCA Vol 4 i1 2025 webmag - Flipbook - Page 11
2025 | Volume 4, Issue 1
Con昀椀guration of a Customized ChatGPT for
ISO/IEC 17025: L-Squad
L-Squad, a customized ChatGPT model, was
con昀椀gured using advanced ChatGPT options with a
“Plus” account and the “Create” tool. This model was
trained with the following guidelines:
1. Objectives and Scope:
Function as a specialist in ISO/IEC 17025:2017
with extensive experience.
Provide guidance on implementing and complying
with the standard’s requirements, emphasizing
consistent operation, impartiality, and laboratory
competence.
2. Accuracy in Requirements:
Do not label any requirement as “mandatory”
unless explicitly stated in the standard.
Clearly distinguish recommendations or best
practices from mandatory requirements.
Avoid presenting “preventive actions” as
mandatory, as ISO/IEC 17025:2017 follows a riskbased approach without explicitly requiring them.
3. Documentation and Procedures:
Specify whether a document or procedure
is explicitly required by the standard (citing
the relevant clause) or a non-mandatory
recommendation.
4. Response Format:
Maintain a formal and professional tone.
Provide concise responses with speci昀椀c
references to the standard.
5. Out-of-Scope Topics:
Address only inquiries related to ISO/IEC
17025:2017.
If asked about unrelated topics, clarify that the
model is limited to ISO/IEC 17025.
11
L-Squad has evolved through ten iterative versions,
incorporating updates that form part of its knowledge
base to deliver precise and contextualized responses
regarding the application of ISO/IEC 17025 in the
management systems of testing and calibration
laboratories. Key documents integrated into its
con昀椀guration include ISO/IEC 17025:2017, ILAC
policies such as ILAC P14:09/2020, ILAC P10:07/2020,
ILAC P9:01/2024; ILAC guides such as ILAC
G8:09/2019, ILAC G24:2022, and ILAC G17:01/2021;
as well as supplementary documents like the AENOR
Pack UNE EN ISO/IEC 17025:2017, ISO 10012, ISO
10009, and Eurachem guides.
Additionally, actions were taken during each update
to re昀椀ne responses based on user feedback. Its
latest instruction was enhanced using ChatGPT o1,
“Uses advanced reasoning,” which strengthened its
advanced reasoning capabilities and accuracy in
normative contexts. Training through reinforcement
learning allowed L-Squad to optimize its responses
through iterative user feedback, ensuring a closer
alignment with user expectations and the standard’s
objectives.
Evaluation of Comprehension Level
To assess the generative AI’s comprehension of the
ISO/IEC 17025 standard requirements, a 40-question
test was developed. This test was divided into two
sections: one consisting of true/false questions and
another with multiple-choice questions that had only
one correct answer. The questions were categorized
into three levels of comprehension:
• Literal comprehension: Focused on the ability to
identify explicit information in the text, evaluated
through multiple-choice and true/false questions
(35 questions). Maximum expected score: 35.
• Inferential comprehension: Assessed the ability
to deduce implicit information from the text,
requiring contextual analysis (昀椀ve questions),
evaluated through multiple-choice questions.
Maximum expected score: 5.
• Criteria-based comprehension: Involved justifying
answers based on critical judgment and alignment
with the standard, using the same 昀椀ve inferential
questions to evaluate this dimension. A speci昀椀c
scoring system was applied, ranging from 0 to 2
points, to evaluate whether the responses were
correctly justi昀椀ed with a direct relationship to
the ISO/IEC 17025 standard. Maximum expected
score: 10.
