We gotta get past RAG and start doing more GAR - GenerativeAI Augmented Retrieval Make search work better with LLMs: - Query understanding / expansion with few-shot LLMs - "SEO"ing our content for our search with LLMs ...

OpenUK warns of the risks of not understanding what openness and open source mean in the context of AI, enabling bad actors to abuse the benefits of open source and to 'open wash'. openuk.uk/press-posts/uk… #opensource

Just to note that most of the techniques listed here have existed for decades, are regularly used in 'traditional' search engines and are nothing to do with AI.

Here's how our client Silverchair have been letting publishers experiment with different AI models etc., using the RAG evaluation system we developed for them as a basis opensourceconnections.com/case-studies/s…

As has ever been the case...policies & action on content authoring, data quality, expiry of old documents, removal of duplicates etc. can make a huge difference to the perceived performance of a search engine.

Remember the Call for Papers for Haystack EU 2024 ends soon, please submit your talk by 9th August (Friday!) opensourceconnections.com/haystack-eu-20… - tell your story at the only conference focused on search relevance! #haystackconf

Looking forward to being part of this series! I spoke with Nicolay about the importance of measurement in building search & AI

You're all putting the final polish on your Haystack Europe talk submissions, right? Right? haystackconf.com

Want to learn how to tune #OpenSearch for better relevance? Find out in a special one-day intensive course at OpenSearchCon US - discounted by 50% ! Space is limited so don't wait - opensearch.org/events/opensea…

Sometimes, the obvious must be studied so it can be asserted with full confidence: - LLMs can not answer questions whose answers are not in their training set in some form, - they can not solve problems they haven't been trained on, - they can not acquire new skills our

Happy to announce Eric Pugh and I will be part of @Algolia's Developer Day on Oct 2nd! algolia.com/devcon/

Relevance in search is subjective, so how can we measure it? Often people focus on the easy to measure stuff like search speed. Or they go by complaint and only tackle areas, where the customer's are the loudest. To evaluate search systems, you need multiple methods working

Working on the talk schedule which should be published very soon....amazing lineup this year! Grab your tickets before they go, space is limited.

Many teams don't do this simple search relevance life hack, before endeavoring on that 2 year rewrite-to-new-cutting-edge-thing-initiative look at your lowest performing queries and fix problems (content, ranking, ...?) that prevents it from being successful

*Heads up* - for Haystack EU the hotel deal is available only until September 3rd - so please book today or tomorrow if you want the conference hotel deal. haystackconf.com/eu2024/travel/

The draft talk schedule for Haystack Europe on Sep 30/Oct 1 is published! haystackconf.com/eu2024/ - get your tickets today to join us to hear about subjects including Measuring Search, Scaling, E-commerce search, RAG, LLMs, visual search, ranking & LTR, vector quantization..[1/2]

...with a keynote (and a chance to get your book signed) by Trey Grainger, lightning talks, Women of Search & more! Remember we have limited space for the in-person event - get your tickets at haystackconf.com

Two training opportunities to level up your skills in search & AI: 1. at OpenSearchCon on Sep 26 join a one-day intensive Think Like a Relevance Engineer course for OpenSearch - *half the usual price* thanks to AWS' support opensearch.org/events/opensea… ... [1/3]

..and just after Haystack Europe let René Kriegler take you down A Pathway to #RAG (conference entry & lunch included) in Berlin on Oct 2nd opensourceconnections.com/training/a-pat… #AI #LLM ...[2/3]

René Kriegler ..remember all our Think Like a Relevance Engineer trainings for #solr #elasticsearch #opensearch plus Hello Learning to Rank are also available on-demand as self-led courses - the same materials but pre-recorded, with quizzes & examples opensourceconnections.com/training/ [3/3]