Author of the best-selling book Marketing to Mindstates, 2x CEO, and Behavioral Market Research Expert.
#TIL: If your job is to provide 'actionable' consumer insights that drive brand sales, using ChatGPT4.0 and other LLMs to analyze your data is NOT very useful. Faster Insights: Yes Better Insights: Sometimes Actionable Insights: Not if you believe in behavioral science LLMs are not trained to apply behavioral science-based models and concepts to data analysis, so the insights and actionability you get out of GenerativeAI are mediocre at best... and misleading at worst. Natural Language Processing (NLP) models can be trained to apply behavioral science principles in data analysis, resulting in highly actionable insight and recommendations. Tomorrow I'll show you a comparison between the two.
Ed Powers
7mo
Actionable, in my view, implies causal, but AI is based on prediction, which doesn’t depend on causation. Businesspeople must examine behavioral effects of their “treatments” using proper methods, and AIs are incapable of reasoning for them—that still requires a human brain.
This very much resonates with our own experience with training LLMs and MMLLMs (multimodal LLMs) William Leach . They do good when it's about what is stated preferences and similar conscious responses (it is after all a language model wmyoure using with an LLM...) The big difference comes when we try to make LLMs work on unconscious responses and behaviors. Then the models fall short. You can still use LLMs but only if you do it in a particular way. The best option thus far seems to be first to train ML models to predict unconscious responses, and only then train an MMLLM to interpret these results and make them actionable.