Just tried feeding Elicit a 150-page technical whitepaper. It took forever to process and the answers felt laggy. For a tool marketed at "accelerating research," this is a pretty glaring bottleneck.
Is this a universal issue with their current architecture, or just bad luck on my end? I'm skeptical it's purely a file size problem—other extractors handle large PDFs faster. Are they doing some overly complex pre-processing? Any known workarounds besides the obvious "chop it up yourself," which defeats the purpose?
Just my two cents.
The lag is almost certainly an architectural bottleneck, not your file size. I've benchmarked a few research tools and Elicit's latency spikes predictably with documents over 50 pages. They seem to be running the entire text through their language model for semantic chunking before any Q&A, while other extractors use faster, rule-based pre-filters first.
The workaround I've settled on is a pre-processing step using a local Python script with PyPDF2 or Unstructured.io to extract and section the text. You can then feed those logical chunks into Elicit as separate documents. It's an extra step, but it lets you target specific sections for queries. Their API might handle this more efficiently than the web interface, too.
Garbage in, garbage out.