The State of Gen AI insiders Report FINAL 0724 - Flipbook - Page 9
Companies should start by mapping their large
language models around their proprietary data
interactions, which correlations and
PROPRIETARY
DATA EXAMPLE
suggestions that are surfaced by the LLM, user
BloombergGPT is Bloomberg’s
action from the suggestion and the resulting
50-billion parameter large language
domain-specific business outcome.
model, purpose-built from scratch
set. This data set would include user
PROPRIETARY
DATA
FLYWHEEL
for finance, using a diverse set of natural
This data is constantly leveraged to tune the
language processing (NLP) tasks within
LLM and creates business defensibility. This
the financial services industry. This
iterative process has a net positive impact
model assists Bloomberg in improving
on the quality of generative output - models
existing financial NLP tasks, such as
improve over time, creates new opportunities,
sentiment analysis, named entity
and enhances user conversations beyond the
recognition, news classification, and
first interaction.
free-form search.
Tools that can help companies with building their proprietary data flywheel include
large language models, synthetic data, vector databases and data clean rooms.
THE STATE OF GENERATIVE AI INSIDERS REPORT
9
