A company is implementing a large language model for two distinct applications: summarizing complex legal documents where precision is critical, and generating creative marketing slogans where some ambiguity is acceptable. They are considering two input sequence compression techniques to reduce inference costs:

• Technique 1: A rule-based method that removes a predefined list of 'filler' words (e.g., articles, prepositions, and common adjectives) from the input text.
• Technique 2: A method that uses a smaller, secondary model to generate a concise summary of the original input, which is then fed to the main model.

Analyze the suitability of each technique for both applications. In your analysis, compare the potential impact of each technique on computational savings and the preservation of essential information.