The following article is interesting:
https://wadetregaskis.com/a-short-update-on-getting-answers-on-the-modern-internet/
In context of this, it would be helpful to consider the following:
The proposed feature is an advanced personalization system for Kagi search results, allowing users to set their own ranking preferences for sources. When no user preference is present in context of a search, then better defaults could be utilized to dial up the quality to be better than even what the author considers the highest quality results as per Bing presently. This feature is designed to prioritize information based on user-defined relevance, ensuring that search results align with individual preferences and needs. Users can specify which sources they trust and want to see more often (this functionality is already in place by Kagi), and which sources they wish to see less or not at all. This will significantly enhance the user experience by providing more relevant and personalized search results, reducing the frustration of sifting through unwanted content, and making the search process more efficient and tailored.
Users should interact with the enhanced custom source prioritization feature seamlessly, with minimal manual adjustments (eg, no new settings / preferences). The aim is to make this functionality as natural and intuitive as possible, leveraging existing Kagi capabilities without adding complexity. Here are detailed scenarios:
Implicit Source Prioritization:
- Example: A user searches for "latest developments in quantum computing." Kagi, recognizing the user’s implicit preferences (prioritization settings), frame context ("small web" vs other lenses) - and (less important; but history is likely undesirable for many/most users, and thus I think "lenses" provide the proper focus here instead) perhaps historical search patterns, automatically prioritizes sources such as recent research papers, technical blogs, and niche scientific publications over mainstream news or generalist sources.
- Implementation: The search algorithm dynamically adjusts based on users' source prioritization settings and lens choice to fine-tune initial presentation, particularly from the AI. Again, this process happens in the background, requiring no additional input from the user.
Enhanced Default Settings for New Users:
- Example: A new user searching for "is the order of a group the lcm of the order of its elements?" receives results that are optimized based on generally high-quality, contextually relevant sources, without having to set any preferences manually.
- Implementation: Utilize a refined set of default ranking algorithms that prioritize credible and contextually appropriate sources until the user sets priority and lens preferences. These defaults ensure that even without user customization, the search results are of high quality.
Contextual Relevance Adjustment:
- Example: A user searching for "climate change research 2024" will see results that Kagi has learned are more relevant based on previous searches related to scientific studies and cutting-edge research. This could include sources like research institutions, academic journals, and authoritative scientific blogs - even if it is contrary to mainstream media preferences and narratives.
- Implementation: The search engine uses prioritization choices and lenses to prioritize results and AI summaries. This dynamic adjustment improves result accuracy and relevance.
AI-Enhanced Summaries:
- Example: Basically already mentioned, but when a user searches for "latest advancements in renewable energy," Kagi presents an AI-generated summary at the top of the search results page, highlighting key findings.
- Implementation: By using prompt injection from the user's prioritized sources and the current lens setting, the AI leverages the user's implicit preferences and search history to generate summaries that are likely most relevant to the user based on both preferences and knowledge, providing quick and accurate insights without additional user input.
Automatic Feedback Integration:
- Example: A user frequently interacts with certain types of sources or specific websites. Kagi automatically learns from these interactions and adjusts future search results to reflect these implicit preferences, showing more of the user's trusted sources and less of the deprioritized ones. Give the user a thumbs up / thumbs down for the AI / Summary sections to provide quick feedback input.
- Implementation: Incorporate machine learning algorithms that analyze users' like/dislike of results when given, and/or click patterns and dwell time, to continually refine and personalize search results. This learning process is entirely automatic, enhancing the user experience without requiring manual feedback except for the option of acceptation of a thumbs up / thumbs down.
