I built a RAG pipeline for SEO use cases with @zilliz_universe and @DeepnoteHQ . Sharing a detailed step-by-step guide on why and how to set up your own RAG (with 0 cost other than the LLM API calls) to do SEO smarter not harder

This also means that by tracking the #:~:text fragment clicks with a GTM Custom JS variable, one can get a pretty clear picture on how much clicks are coming from AIO links.

Fun fact. The new Gemini embeddings model has a task type named RETRIEVAL_QUERY, along with some of the other standard ones like RETRIEVAL_DOCUMENT. From the documentation: .Embeddings optimized for general search queries. Use RETRIEVAL_QUERY for queries.

In case anyone is going through the same thing - I've found that the pseudonym is a good approach 🥸. I think less about:. "is this going to hurt me" or "maybe this is completely wrong". And more about: "how to make this more useful" and "showing what I did so others can use it".

RT @googleaidevs: 📢 The Gemini Embedding text model (gemini-embedding-001) is now generally available in the Gemini API via Google AI Studi….

. Google has decided to publish less research papers and this is a big example of that policy change. Is it because this innovation is so big they want to keep this as a competitive advantage?. No it's cause @metehan777 and @KorayGubur been reverse engineering everything 🤣.

Google out with first Gemini embedding model API.

RT @visualizevalue: "Life is trying things to see if they work." – Ray Bradbury

So it might be interesting to compare in the real world (SERPs) how well the single vs. multi vector approach does especially with huge documents. Regardless of the difference in speed.

". unlike single-vectors. FDE’s are guaranteed to approximate the true Chamfer similarity to within a specified error. MUVERA is guaranteed to find the most similar multi-vector representations."

Amazing write-up to try to mimic a MUVERA-ish approach!. And - as I'm reading about MUVERA one fact seems to be hiding in plain sight within the original Google blog announcement, which is that MUVERA should also be INCREASING PRECISION of the retrieval?.

402 Payment Required. Cloudflare going all "freelance contractor" on your ass 🤣.

Please tell me this already exists. Remember the "consensus vs. contrarian" discussion in SEO?. How interesting would it be to revisit it but from the present context - for example from a purely numerical/stats view of cosine similarity between documents vs ranking ranges.

I do not subscribe to the classic Gartner hype cycle of tribal SEO bullshit. Not hiking along to the Peak of Inflated Expectations. Not digging your Trough of Disillusionment. Just adopting early and learning about what will be useful long after we hit the Plateau of Productivity

RT @martinibuster: Google published details about a new multi-vector retrieval algorithm that's one of those milestone announcements. All u….

RT @RyanJones: Got my own python implementation of google's mumvera working based on the research paper logic and existing colBERT models.….

RT @lilyraynyc: Just some of the crap I read on LinkedIn these days. Make it stop 🥵

Time to repost for all you SEO nerds: ChatGPT deep research can't do this for you!. Spent a lot of time to scale up Information Gain for highly competitive SERPs (patent: . Useful for AI overviews but also predictive content gaps.

SEOs rn when asked by stakeholders about AI.

RT @searchmartin: SEOs panicking about AI: