- GWDG AI summit in September 2025
- various updates of open models
- SOLR has dense vector search: , Neural Information Retrieval
This needs a way to turn metadata into embeddings (one-time operation).
At query time, the embedding needs to be calculated for the query and then could be matched against the index/dense vector fields.
Questions:
- how different would this be from more like this?
- computational effort at query time
- how to evaluate quality
Expose the catalog as a research tool to a MCP host. That could be a desktop application.
Who uses a desktop LLM client?
May help, if adapters could be implemented to different search engines and catalogs. "VerbundMCP"
A paradigm to combine search over documents and LLM.
- ragweed, mediarep example, local LLM, haystack
What is huggingface?
The key is to note that the benchmarks used to test LLMs—such as AP exams—are also those used to test people. However, this raises an implication: these benchmarks are only valid tests if LLMs misunderstand concepts in ways that mirror human misunderstandings. Otherwise, success on benchmarks only demonstrates potemkin understanding: the illusion of understanding driven by answers irreconcilable with how any human would interpret a concept. -- https://arxiv.org/pdf/2506.21521
...
... benchmarks designed for humans are only valid tests for LLMs if the space of LLM misunderstandings is structured in the same way as the space of human misunderstanding ...
