I have been working on a grounded question answering bot using , fancy prompting and google search for a while now!
today is releasing a bunch of open source project using LLMs, and my project is one of them!
this video explains :)
youtube.com/watch?v=DpOQpC🧵
Conversation
Replying to
Read the full blog here: txt.cohere.ai/introducing-sa
I have been working on this in part because it attempts to address a fundamental problem of LLMs.
LLMs are reliable at creating sensible answers to complex questions, but they aren't good at coming up with truthful answers.
1
4
21
They are trained on data from the web, and so pick up statistical correlations between words that make them ok at answering simple and static questions (things like "how far away is the moon from the earth", which has a single and unchanging factual answer).
1
9
However, more nuanced questions or ones that have factual answers which change over time are difficult or impossible for language models to answer.
"Who is the prime minister of the UK" for example is actually hilariously hard to answer these days
1
8
The Grounded QA Bot i have been working on lets you deploy a contextualized, factual, question-answering conversation bot that uses embeddings, prompting, and web search to try to improve on this fundamental limitation of LLMs.
2
13