Curate your own
x.comTreg (@treg_ai) · Post
An open-source toolkit that helps users find every source AI cites and identify the authors behind those sources, while enabling products to get indexed in the sources AI trusts. The tool costs 1 cent per prompt with no subscription requirement and provides access to over 4000 data sources and services. Available as a public repository with partnership opportunities through Cloro AI.
#ai#open-source#product#ai saas#developer
instagram.comReel
IBM's STAIR approach eliminates traditional RAG's chunking, embeddings, and vector databases by instead using a document's table of contents. The model learns to read a query and point directly to the relevant section (the leaf node in the table of contents hierarchy), which is then passed as context to generate answers. Tested on 18 textbooks across 6 domains, STAIR achieved 82.6% retrieval quality versus 76.9% for the DSI baseline, significantly reducing hallucinations. The main limitation is that documents must have clean, well-structured tables of contents; the researchers plan to next develop methods for generating artificial tables of contents and scaling to enterprise documents.
#ai#research#machine learning#ai saas
x.comShengkun Ye (@shengkunye) · Post
A developer announced a free web search and fetch capability for AI agents, positioning it as an alternative to paid services like Exa, Tavily, SerpAPI, and Brave's search APIs which charge around $7 per 1,000 searches. The claim is that agents should have free access to web search similar to how humans use Google for free, with no subscriptions or usage quotas required. The project was made possible through collaboration with Tiny Fish and Monid.
#ai#automation#open-source#developer#product strategy
