zeit.de
ZEIT ONLINE" Launches AI-Powered Beta Search for Articles Since 1946
ZEIT ONLINE" launched an AI-powered beta search tool accessing articles from "DIE ZEIT" and "ZEIT ONLINE" since 1946, generating answers from linked sources, adapting to query type (simple or complex), and using various AI models, including an OpenAI language model, with user feedback crucial for ongoing development.
- What are the limitations of the beta search, and how can users contribute to its improvement?
- The beta search is still under development, with potential issues like unsatisfactory answers or irrelevant results. User feedback is crucial for improvement, and variations in query phrasing may improve results. The AI aims to avoid fabricating information, relying solely on sourced articles, although outdated or mismatched information may occur. Manual time-frame limitation is available for advanced search.
- What is the core functionality and data source of the new "ZEIT ONLINE" beta search application?
- ZEIT ONLINE" launched a new beta search application powered by AI, allowing users to search articles published in "DIE ZEIT" and "ZEIT ONLINE" since 1946. The search uses a language model trained exclusively on articles from these sources to generate answers to user questions, prioritizing relevant excerpts and sorting them by relevance. New articles are added within an hour of publication.
- How does the search application differentiate between simple queries and complex questions, and what technologies are involved?
- The search functionality adapts to user input, offering tailored results for simple searches or detailed answers for formulated questions. Different AI models, including some trained in-house, are combined to process queries, with an OpenAI language model currently used for answer generation. The system weighs articles differently depending on whether the query concerns recent events or a broader timeframe.
Cognitive Concepts
Framing Bias
The framing emphasizes the beta nature of the search and its limitations, potentially downplaying the capabilities and achievements. The focus on potential shortcomings (e.g., inaccurate answers) might disproportionately influence user perception.
Language Bias
The language used is generally neutral and informative, though phrases like "very low" (referring to the probability of hallucination) could be considered slightly subjective. The overall tone is cautiously optimistic but emphasizes transparency about the system's limitations.
Bias by Omission
The description focuses heavily on the functionality and limitations of the search application, potentially omitting discussion of the broader implications of AI-powered news search or the potential biases inherent in the underlying data. There is no mention of how the selection of articles in the database might skew results.
False Dichotomy
The text presents a false dichotomy between simple search queries (yielding only matching hits) and formulated questions (generating an answer). The reality is likely more nuanced, with various degrees of query complexity and response types possible.