Loading video player...
Hi everyone and welcome.
I’m Liam and today I’m joined by
Naomi and Alex from Azure AI Foundry.
Thank you for joining.
I understand, Naomi,
that you’re going to be talking a little bit
about GPT-5.
Can you give me a little bit of the
highlight of that model?
Yeah, absolutely Liam. I’d love to.
So GPT-5 is the latest addition
to the Azure OpenAI family.
It's the first model suite
that really unifies chat
and reasoning capabilities,
and intelligently routed
by our model router
to optimize performance and save cost.
It's really designed for real-world,
high stakes workloads.
Whether that's advanced coding,
it might be deep analytics,
decision making,
or powering enterprise copilots.
That makes a lot of sense,
but are there any areas
where you've found that it
particularly stands out?
Yeah, there's really three big things
I would mention.
First off, is that frontier reasoning capability.
GPT-5 really excels
at that multi-step logic
and contextual problem solving.
It can reason over large, complex inputs.
Think about thousands of lines of code
or hundreds of pages of documents
without losing that context.
It has also high performance generation.
It's not just about thinking,
it's about delivering
usable and production-ready output code
that runs, analyzes, and stands up
to scrutiny and language
that matches your brand tone.
Lastly, I'll mention
cost-efficient scaling.
So thinking about
the Azure AI Foundry model router,
we can think about customers
can automatically get that right
GPT-5 variant for their task.
Whether that's GPT-5 reasoning
for deep analysis,
they might use GPT-5 mini
for faster turnaround,
or GPT-5 nano for lightweight calls.
Well, that's obviously very compelling,
but Alex, let's be real.
People can get GPT-5
at places outside of Azure.
Why should organizations
be looking at Azure AI Foundry
for this model?
Absolutely Liam.
So enterprises
choose Azure AI Foundry
for the trusted security and compliance.
Also for the seamless integration
they get across the entire
Microsoft stack.
And of course, also the governance tools
that they can use to deploy
GPT-5 responsibly.
And lastly, as Naomi mentioned,
with the model router,
customers get the best balance of cost
and performance.
That makes sense.
Having the trust
and security in their models
is so important
to all of the organizations we work with.
Now Naomi,
you mentioned advanced
frontier reasoning.
I wonder if you can tell me
a little bit more about that and maybe
do you have anything you could show us?
I'd love to.
Of course I'd love to show off
how GPT-5 can assist researchers,
especially in peer reviewing academic work.
So first off, we're going to start off
by prompting GPT-5
to act as a critical peer reviewer.
So here's the prompt that we're using.
First off, act as a critical peer reviewer.
Identify any logical flaws,
suggest extensions, flag any verbosity,
and then provide 20 concrete extensions.
And you can see this sets GPT-5 up to deliver
structured and actionable feedback
that is both practical and thorough.
And what kind of feedback does
GPT-5 actually generate?
So you can see here like
it’s returning a list of suggestions,
each with a clear header
and a concise summary.
So let's look at some of them.
For example,
this paper here doesn't
explain how agents are chosen.
So GPT-5 is suggesting
how to add selection heuristics.
It also recommends
elaborating on success metrics,
in this case to improve clarity
and reproducibility,
and then GPT-5 is also flagging that there’s
a lack of discussion on performance
under large scale scenarios.
Wow, so we're basically getting
technical critique here from GPT-5.
I wonder, does GPT-5 go beyond this?
Like for example,
can we adjust the writing style?
Yeah exactly.
So this is a great question.
So it flags that verbose
or overly technical language,
and it's just suggesting simpler
alternatives for
the mere humans among us.
So you can see here in seconds, GPT
GPT-5 has delivered
20 targeted suggestions
from those structural improvements
to the future research directions.
So with that, of course,
Alex, do we still need humans
to review our papers?
Of course.
So while GPT-5 is a powerful assistant,
human peer review remains
absolutely essential.
So academic research
is going to demand
critical thinking, domain expertise,
and perhaps most importantly,
ethical judgment.
And this is only something
that humans can provide.
Exactly.
I can see this advanced frontier
reasoning being so applicable
to so many different organizations,
not just in research.
It's really great.
Now Alex, we’ve seen how
this can help researchers,
but I wonder,
can you tell us a little more
about how this can help developers,
perhaps to build more advanced applications?
I heard you might have a cool little demo.
Is that something you could show us?
Absolutely.
So Liam, I'm actually using GPT-5
integrated directly into VS Code,
using GitHub Copilot
to generate code for a website
where students can enter different words
like nouns, verbs, adjectives.
And the app is going to generate
both a short story and a matching image.
So what I'm trying to do is create
a fun way to combine
natural language with creativity.
That sounds great.
So how does GPT-5 handle that?
So what we do is we start
with a single prompt.
Here's the prompt.
And the prompt has everything GPT-5 needs
to create the complete application.
So in real world environments
people will typically start
with a simple prompt.
And they're going to use GPT-5
to iterate over that prompt
and make improvements.
That makes sense
because what you're showing here
is quite an extensive prompt,
which is great,
but I can imagine
most people are starting
small a couple of sentences
just to get the idea and start iterating,
is that the right way
of thinking about this?
Absolutely Liam.
So what we have found is that GPT-5 does
quite well at following instructions.
So as you can see in this prompt,
the more details
you give the better in helping you create
exactly the application that you want.
This is great.
So can you show me a little bit more
about what it looks like
once it actually builds the application?
Absolutely.
So let's take a look here at the code.
So GPT-5 is going to instantly produce
a complete HTML page.
We see the embedded JavaScript.
It's going to handle the form inputs,
dynamically build a story using the words,
and then also call an image generation
API like Azure OpenAI Image 1
to visualize the user inputs.
Wait a minute,
we're using GPT-5
to create an application
that's going to use GPT-5
as well as image gen.
That's pretty cool.
I like that.
Absolutely,
and it's all available in Azure AI Foundry.
Let's take a look, Liam
here at the website of what's
generated by this code.
So as you can see here,
the user, the student,
can enter in the various words.
And they can also see the story
and the corresponding image.
Wow, that's amazing.
Just from a single prompt.
Now Naomi, like this is cool
and amazing in the education space,
but I can actually imagine
this might be applicable in other areas.
Is that right?
Yeah, that's exactly right.
So what you've seen here
of course with education,
but we could use this
in all kinds of enterprises
and industry use cases.
And so being able to use perhaps
brand imagery maybe, or recreating,
you know, different portfolios,
or maybe you've got product imagery
that you want to do as part of this too,
this is a great way to be able to use
all of that power of GPT-5.
Amazing.
That's a great example of how GPT-5
makes coding more intuitive
through natural language.
But coding is just one side of the story.
Naomi, I understand through GPT-5
there's something called tool calling
as well as free-form tool calling.
Can you tell me a little bit more
about what that means
and why that's so important?
Yeah, I'd love to share that with you, Liam
and most importantly, to show you it.
So first off, let’s look at one of the
most developer-centric features
that we have in GPT-5.
So instead of rigid JSON, GPT-5
can actually call tools with natural text
like raw Python scripts,
or SQL queries, or configs.
And that means faster integration
and more intuitive workflows.
So unlike traditional tool
calling that leverages JSON,
the free-form makes it more natural language
that allows us to really extend
the types of tools
such as calling Python code
or SQL, as you mentioned.
Is that right?
That's exactly right.
So no wrapper and there is no friction.
So let's check it out.
Let's imagine that we are inventory specialists
at a national retailer.
One of our daily tasks is to generate
those weekly revenue summaries.
So normally this means
I’ve got to write SQL code,
I’ve got to run it,
I’ve got to export the data.
And then I’ve got to format it with Python.
So it's time consuming
and it requires
a lot of different handoffs.
So with GPT-5’s free-form tool calling,
we can really make this an easy way
to automate that entire workflow.
So the model is going to generate that SQL,
it's going to call a tool to execute it,
and then it's going to call another tool
to format the results
and it’s doing it all in one loop.
No JSON and no switching tools,
just natural language.
So now we're defining two tools here,
So SQL Execute Lite
basically helps us here.
So we're going to define two tools
sql_exec_sqlite runs
that SQL and returns results in CSV.
code_exec_python is going to run our Python
and return our output.
So I'm going to ask GPT-5 here
to generate that SQL
to create a sales table
and that compute revenue.
And then I'm going to ask it
to go run a conversation loop.
We're going to use that loop
to basically send the prompt to GPT-5,
detect that there's a tool called,
execute the tool,
and then send that output back to GPT-5
and continue until it's done.
So as you can see here,
GPT-5 has returned a neatly formatted
summary of all of those product
revenues that we need,
all via orchestrated toolchains.
Well that's amazing.
So not only did GPT-5 figure out
which tools it needed to use,
but it created
all of the natural language text
to actually call those tools
and execute it on its behalf.
That's really amazing.
I can imagine how this could open it
up for all kinds
of different types of tools
within applications.
Exactly.
So there's really three big ones
that we can look at,
so there's no JSON that's been required,
you can see that's natural output,
and then there's
multi-tool orchestration.
All of this makes it
a very developer-centric tool.
So we've simplified scripting,
we sped up all of that reporting,
and we've enabled richer
and more intuitive interactions
into the data.
GPT-5 maintains context
across all of these steps,
so it can chain
those tool calls intelligently,
and it makes it perfect
for dynamic automation
and multi-tool orchestration.
This is really a game changer
for developers in any industry.
It's absolutely true.
I mean, this is going to be applicable
to so many different
organizations and developers.
Naomi and Alex, amazing demos.
Before we wrap up,
what should AI practitioners
and developers take away from today?
Yeah, so GPT-5 is so much more
than just a model.
It is a toolkit for reasoning,
for generation and orchestration.
And Alex, how can people get started?
So the best way is to start small.
Pick one high-impact use case
and then try GPT-5
on Azure AI Foundry.
Naomi and Alex,
thank you for walking us through this.
And thanks to all of you for watching.
And go start building with GPT-5 now.
Discover how GPT-5 in Azure AI Foundry, is transforming the way developers, researchers, and enterprises build intelligent applications. In this demo, Microsoft Azure AI product experts walk you through three powerful real-world use cases and showcase GPT-5 in action—building apps, reviewing research, and automating code using natural language, multimodal inputs, and frontier reasoning. Build your smarter, faster and more creative agents and apps with GPT-5 in Azure AI Foundry now! https://msft.it/6055sgDPZ