EmbeddingGemma 2: an open, lightweight multimodal embedding model
deepmind.google·6d ago
AI Engineering
Hacker News·Sep 2·5 min read
Three sites made 215,128 "best software" pages for AI. Perplexity cites them
TL;DR
Three newly established sites generated over 215,000 machine-created 'best software' pages across 380 categories, raising concerns about the quality of sources. A significant portion of cited domains ranked poorly or were not in the top million.
✦ Why It Matters
Evaluate the sources of software recommendations to ensure quality and reliability in your decision-making.
Key Takeaways
- Scrutinize AI-generated recommendations for source credibility.
- Be aware of the prevalence of low-traffic domains in citations.
- Consider the recency of cited domains when evaluating recommendations.
- Investigate the ownership and purpose of sites generating software lists.
Full Summary
In a study, two AI models, perplexity/sonar and perplexity/sonar-pro, were tasked with identifying top products across 380 software categories, yielding 7,534 citations. Notably, 59.8% of these citations pointed to domains ranked worse than #100,000 in the Tranco top-1M list, with 23.4% from domains not in the top million.
Three sites, registered between December 2023 and May 2024, collectively published 215,128 machine-generated pages, raising concerns about the quality of information. The analysis revealed that many cited domains are relatively new, with a median first capture date of 2020.
The findings suggest that the AI models may rely heavily on less reputable sources, which could affect the quality of software recommendations.