Transforming the Search Industry with LLM Using Source Frontier
summary
Claims a production information retrieval layer called Source Frontier that already stores tens of billions of vectorized text segments, integrates with an LLM to generate answers that cite multiple sources, and is fine tuned with about 10,000 fine-tuned examples bootstrapped from model outputs. The pipeline runs cheaply—on one or two servers for low-latency requirements—reduces hallucinations by forcing evidence-backed atomic facts, scales incrementally (plans for ~100 billion answers), and reports usable results in demos.
AI-written summary · about
chapters (10)
- 0:00LLM impact on search
- 1:07Y answers background
- 3:19LLM capabilities and limits
- 4:21LLM hallucinations and validation
- 7:14Source Frontier tech
- 9:26Source Frontier demo
- 10:56LLM in Source Frontier
- 12:39Advantages of combining LLM with Source Frontier
- 14:25Future goals and scaling
- 14:51Q&A
related talks
24mConversational Search with LLMs
49mLLM—Search Engine Killer?
51mQuery Understanding With Large Language Models: Techniques and...
21mTransforming search and discovery using LLMs
46mLLM—Search Engine Killer?
15mSearch Engineering Reimagined: Leveraging LLM in ML Workflows
01h 30mUsing LLMs to improve your Search Engine
47mGive Your LLMs a Left Brain