Please wait while the transcript is being prepared...
0:00
Hi. My name is
Alejandro Lopez-Lira.
I'm an assistant professor of
finance at the
University of Florida.
In this lecture, we're
going to be covering
large language models
and financial analysis.
0:15
First, we're going to start with
a brief introduction to
large language models.
0:21
Large language models are
artificial intelligent
systems trained to process,
understand, and generate
human language.
Examples include ChatGPT, GPT-3,
BERT, Transformer models,
Claude, and Bard [now Gemini].
How do these models work?
These models are built on
deep neural networks, mimicking
aspects of the human
brain function.
These are trained on
vast corpuses of texts
basically as many
texts as you can get,
if you can get the
whole Internet,
it's better, and they learn
language patterns and nuances.
These large language models have
advanced capabilities.
For example,
they have contextual
understanding.
They can grasp subtle
meaning and context in text.
Moreover, they have
a great generation.
They can produce coherent
and contextually
appropriate language.
These large language models
represent a significant
leap compared to
the basic natural language
processing models
because they have
a sophisticated
understanding and
a great capability of
generating language.
1:28
Let's cover a little bit of
history of the evolution
of large language models,
I promise it will
not be too much.
From the 1950s to
the early 2000s,
there was not a lot of progress.
There were initial
experiments with
natural language
processing techniques,
including conceptual
and rule-based system.
Most of these were
hard-coded rules and most of
the other advancement was due to
some very basic
statistical models.
However, thanks
to deep learning,
there were significant advances
starting in the 2010s.
For example, one key milestone
was the development
of the BERT model.
This was a major
breakthrough in 2019.
This was introduced
by Google and it set
new standards in natural
language processing benchmarks.
Most of these recent
models like ChatGPT and
Claude and Bard are
direct descendants from this
type of model called BERT.
This model was quickly
introduced into
Google search and it
started influencing
mainstream applications.
Now beyond BERT,
especially lately, there has
been a significant progress
towards larger and
more capable models,
so it turned out that most of
the key to success is just
having larger datasets and
having more complex models.
