Text Embeddings Get a Bigger Disk

They split the file into chunks by page the match your query but cosine similarity if the answer is bad so they fill their query full of keywords expecting it to match so they switch to `text-embedding-3-large` and get a bigger disk

A clip of El Risitas laughing uncontrollably during a television interview is overlaid with subtitles explaining a progressively absurd text-embedding and vector-search workflow. The punchline is that switching to a larger embedding model supposedly produces a “bigger disk,” making the technical explanation sound increasingly ridiculous while his infectious laughter sells the joke. EN: A clip of El Risitas laughing uncontrollably during a television interview is overlaid with subtitles explaining a progressively absurd text-embedding and vector-search workflow. The punchline is that switching to a larger embedding model supposedly produces a “bigger disk,” making the technical explanation sound increasingly ridiculous while his infectious laughter sells the joke.

• @gentoomemes / 11070 2024-05-03