🔒Открыть пост
и посмотреть медиа
и посмотреть медиа
How to Interpret a Rating Scale Without Historical Data
UX rating scales are negatively skewed (positive wording + agreement bias). Using SUS distribution as reference: Good = 80% of scale range (4.2/5, 5.8/7), Average = 70% (3.8/5, 5.2/7), Poor = 50% (midpoint). Formula: Target / (100 / (MaxRating−1)) + 1. Or convert to 0–100: (Rating−1) * 100 / (MaxRating−1). Best guesses until you collect your own data
The intelligence revolution won’t be televised — it will be automated over a longer arc
The Intelligence Revolution will take a decade, not 18 months. Like the Industrial Revolution, success requires seeing the whole system, redesigning the process, and offering workers a deal worth accepting. Most valuable work happens in the unmapped "white space" (handoffs, collaboration). Before deploying AI, map the work and redesign the social contract—workers need a reason to accept change
NNG: Small by Design - The Strength of Lean Design-System Teams
Lean design-system teams, when strategically planned, can move faster, prioritize sharply, and scale impact beyond their size
AI: Can AI Detect Usability Problems Like Researchers?
ChatGPT (31%) and Gemini (57%) tested for reliability in finding usability problems (human benchmark 47%). Gemini's reliability was good, ChatGPT's fair. But agreement between the two AIs was low (28%). Reliability isn't accuracy — next step: compare to human evaluators. Different LLMs see different problems
Opinion: The only winning move is not to play
AI in user research removes human meaning-making and leads to average results. Tooling platforms sell AI as a replacement for researchers. The red line: conceding what researchers are uniquely good at to a bot. "I didn't enter this field only to not do the job."
@uxdigest
