LLM training and user needs

Stub

LLM training and user needs is a working area for examining how current large language models often reproduce a weak, behaviourist or product-centric understanding of user needs.

The problem is not only that an LLM may write a bad user need. The deeper issue is that the model may lack a robust civic, strategic and semantic model of what a user need is for.

Working problem statement

LLMs often collapse user needs into tasks, desires, user stories, product features, content requests or observable behaviours.

That makes them risky in complex service and civic design work, where a useful need may need to preserve:

• the person’s underlying goal
• the emotional and social context
• structural constraints
• civic obligations
• institutional legitimacy
• equity and access conditions
• the distinction between need, solution and delivery artefact

Fault map to develop

This entry should later explore common failure modes such as:

• definition collapse
• behavioural simplification
• solution embedding
• strategic disconnection
• contextual elision
• ontological absence
• context rot across long conversations

Link to Civic Design Intelligence

This is lightly connected to Civic Design Intelligence because the system depends on AI being able to handle user needs, civic needs, evidence and decisions with more semantic discipline than generic LLM defaults usually provide.

Corrective protocol ideas

Possible corrective moves to develop later:

• de-solution every input
• verify fit across civic, external-user and internal-operational layers
• audit for needs erasure
• require multiple concept logics rather than optimising one assumed solution
• monitor context loss in long analysis sessions

To develop later

• review the Needs Simulation material
• decide whether “context rot” should be its own entry
• separate LLM failure modes from human/institutional failure modes
• turn the corrective protocol into a tool or checklist
• add examples only after review