Boost Your LLM Performance with Vast Knowledge
In today’s highly competitive field of AI, we enable you to step up in the quality of your AI product using our knowledge Source Frontier system of tens of billions of answers to open domain questions as grounding to your LLM. Source Frontier system provides direct, concise answers to factoid type of question along with longer texts and multiple source links where this information occurred on the internet. In this way, it recalibrates LLM’s generated answer, minimizing hallucinations and enriching answers with more facts.
Results of different LLMs
Research showed that the Source Frontier makes a significant performance boost in knowledge based tasks for commercial LLMs, and even makes smaller models (~3B parameters) to overperform GPT-4.
| Model name | Base Accuracy | Boosted with Giga answers | Boosted with Tera answers | ||
|---|---|---|---|---|---|
| Accuracy | Improvement | Accuracy | Improvement | ||
| Flan T5 XL | 15 % | 52 % | + 242 % | 56 % | + 274 % |
| Phi 2 | 34 % | 56 % | + 65 % | 62 % | + 81 % |
| Mistral 7b | 37 % | 55 % | + 47 % | 62 % | + 66 % |
| Llama 2 7b | 43 % | 55 % | + 28 % | 62 % | + 44 % |
| Llama 2 13b | 43 % | 60 % | + 39 % | 64 % | + 50 % |
| GPT-3.5 | 56 % | 62 % | + 10 % | 69 % | + 23 % |
| GPT-4 | 61 % | 61 % | + 0 % | 68 % | + 12 % |
Source Frontier
In contrast to common RAG (Retrieval-Augmented Generation), Source Frontier returns shorter and direct point facts for LLM context, which gives more possibility for results customizations with different sources and reduces expenses by decreasing number of tokens in context input.
A sound system that comes from more than one source and is routed through two or more speakers that surround the listener
Basically, stereo means a sound system that comes from more than one source and is routed through two or more speakers that surround the listener.
Which means 'solid' or three-dimensional
The word 'stereo' comes from the Greek word stereos, which means 'solid' or three-dimensional.
A ‘binaural’ form of listening
This comes from an effect known as ‘stereo’ or in human perceptual terms a ‘binaural’ form of listening.