The State of Data & AI - Trevor Jones

Summary

The video script discusses the evolution and impact of AI, particularly generative AI, in the business landscape. It highlights the importance of demystifying AI and understanding its true potential beyond the hype. The script covers the dramatic events surrounding Open AI and Microsoft, the competitive AI space, and the role of data in training AI models. It emphasizes the need for a balanced approach to AI's commercialization and safety, and the integration of AI with established data science and engineering practices for valuable business outcomes.

Takeaways

• 🧠 AI Demystification: The script emphasizes the need to demystify AI, clarifying that it's not magic and builds upon previous technological advancements.
• 📈 Generative AI's Role: It highlights generative AI's capability to capture signals from unstructured data and combine them with structured data sources, enhancing data utilization.
• 🤖 AI in Business: The speaker stresses the importance of using AI in conjunction with established data science and engineering practices to serve valuable business cases.
• 🔄 Open AI Drama: The script recounts the dramatic events involving Open AI, including the dismissal and subsequent return of CEO Sam Altman, and the shift in board composition.
• 🤝 Microsoft's AI Alliances: It discusses Microsoft's strategic moves in AI, including its relationships with Open AI, Mistral AI, and Inflection AI, showcasing its broad engagement in the AI space.
• 💡 Chip Makers' Influence: The script underscores the pivotal role of chip manufacturers like Nvidia and AMD in powering AI training and usage, highlighting their significant market positions.
• 🌐 Cloud Vendors' Competition: The document outlines the competitive landscape among cloud platform vendors—Google, AWS, and Microsoft—as they vie for dominance in the AI-driven cloud market.
• 📊 Investment Surge: The script notes the substantial investments being made in AI by both private companies and governments, indicating a massive influx of capital into the sector.
• 🛠️ Technological Mosaic: AI is part of a larger technological ecosystem that includes sensors, 5G, robotics, and biotechnology, contributing to a rich tapestry of innovation.
• 📚 Realism in AI Impact: There's a call for realism regarding AI's macroeconomic impact, suggesting that while significant, it may not be instantaneous and requires a long-term perspective.
• 📈 Data Centrality: The script concludes by reiterating the fundamental role of data in AI, stating that data is both the foundation for training models and the key to leveraging AI for business success.

Q & A

What was the main aim of the AI and data update event in London on October 19th?

The main aim of the event was to demystify AI, helping people discern between the true potential of the technology and the hype that was building around it.

What is generative AI's role in handling data according to the script?

Generative AI helps capture signals from unstructured data sources and combines them with structured data sources, enhancing the tools available for business cases.

What significant event occurred with OpenAI around a month after the London event?

There was a sudden drama where OpenAI's CEO, Sam Altman, was dismissed by the board, but later returned as CEO with a recomposed board.

What was the reported outcome of Microsoft's involvement with OpenAI during the boardroom drama?

Microsoft was involved in trying to ensure something was preserved from the situation, and it was believed that up to 95% of OpenAI's staff would move to Microsoft.

How did Microsoft's strategy evolve post the OpenAI incident?

Microsoft started to diversify its alliances by striking a deal with a French AI company, MISTL, and later acquiring Inflection AI, indicating a push beyond its previous reliance on OpenAI.

What is Nvidia's role in the AI industry as mentioned in the script?

Nvidia is a major supplier of processing power for AI training and usage, with around 85% of the market share, and has developed a software platform highly valued by model developers.

What is the significance of the bipartisan Senate AI working group's report?

The report calls for at least $32 billion per year in non-defense AI innovation spending, starting from 2026, indicating a significant commitment to AI from the U.S. government.

How is the investment in AI expected to impact the technology's development?

The investment is driving rapid improvements in AI models, including longer input context windows, improved reasoning, and the ability to handle multimodal inputs such as text, image, audio, and video.

What is the importance of data in the context of generative AI as discussed in the script?

Data is fundamental for training foundational models and is essential for applying the strength of these models to a company's unique information and customer interactions.

What is the current sentiment towards the macroeconomic impact of generative AI according to the script?

There is a greater realism about the impact of generative AI, with the understanding that while it will have a significant effect, it may not be instantaneous.

How does the script describe the current state of the AI market in terms of competition and innovation?

The AI market is described as highly competitive and innovative, with many significant players and new entrants like MISTL, and it's not solely dominated by large companies.

Outlines

00:00

🤖 Demystifying AI: Separating Hype from Reality

The script begins with a reflection on the AI hype that peaked in October 2023, following the launch of GPT in late 2022. The speaker's aim was to demystify AI, emphasizing that it is not magic and has evolved from previous technologies. Generative AI was highlighted as a tool for capturing signals from unstructured data, which can be combined with structured data. The speaker stresses the importance of using AI in conjunction with established data science and engineering practices to serve valuable business cases, rather than as a standalone solution. The script also touches on the dramatic events surrounding OpenAI, including the dismissal and subsequent return of CEO Sam Altman, and the potential shift of staff to Microsoft.

05:01

🌐 AI's Competitive Landscape and Strategic Alliances

This paragraph delves into the competitive landscape of AI, focusing on the roles of major players like Microsoft, Google, and Amazon in the cloud platform market. It discusses Microsoft's strategic moves, such as its investment in OpenAI and its partnership with the French AI company, Mistral. The speaker also mentions the importance of safety versus commercialization in AI development. Chip makers like Nvidia and AMD are highlighted for their crucial role in providing the processing power for AI training and usage. The paragraph concludes by discussing the significance of managed analytics platforms like Snowflake and Databricks, which offer cloud-based solutions for data warehousing and engineering.

10:03

📊 Forrester Wave Analysis and Market Presence of AI Models

The speaker presents an analysis based on the Forrester Wave, which evaluates AI foundation models for language. The analysis considers market presence and strategy, with Google positioned as a leader in both. The paragraph discusses the difficulty of categorizing the rapidly evolving AI market and the importance of not underestimating smaller players. The analysis serves to frame the competitive space, highlighting the presence of various companies in the market, including OpenAI, Microsoft, and Mistral, with a note on the peculiar absence of Meta's Llama model from the analysis.

15:04

💰 Investment Surge in AI and its Economic Impact

The script addresses the significant investment in AI by major tech companies like Meta, AWS, Google, and Microsoft, which are expected to invest around 200 billion dollars. It also mentions the US government's spending on defense-related AI procurement and a bipartisan Senate AI working group's call for increased non-defense AI innovation spending. The speaker notes the global nature of this investment, with countries like China also actively participating. The paragraph concludes by discussing the rapid improvement of AI models, focusing on three facets: longer input context windows, improved reasoning, and the ability to handle multimodal inputs.

20:07

📈 The Evolution of AI Models and Their Commercialization

This paragraph discusses the evolution of AI models, particularly the increase in input context windows, which has significantly expanded the amount of text that can be processed. It also touches on the importance of creating sophisticated prompts for these models and the improved reasoning capabilities that come with larger context windows. The speaker mentions the cost associated with using more advanced models, such as OpenAI's 3.5 and 4.0, and how pricing is likely to become more competitive. The paragraph concludes with a historical perspective on the development of technology from the 1970s to the present, emphasizing the layered nature of technological advancements.

25:09

🌐 Realism in AI and Its Integration with Data Science

The final paragraph emphasizes the shift towards realism in the perception of AI's capabilities and impact. It discusses the integration of AI with other data science and engineering techniques, highlighting the importance of using data effectively. The speaker argues that generative AI should be used in combination with existing capabilities to address valuable business goals. The paragraph concludes by reiterating that data is central to the development and application of AI, and that everything can be considered data in the context of AI's ability to analyze unstructured information.

Keywords

💡AI Chat GPT

AI Chat GPT refers to a type of artificial intelligence designed to engage in conversation with humans. In the script, it is mentioned as a significant launch in November 2022 that generated a lot of hype and awareness in the AI field. The video aims to demystify AI, emphasizing that it is not magic but built on previous technological advancements.

💡Generative AI

Generative AI is a subset of AI that can create new content, such as text, images, or music, based on existing data. The script discusses how generative AI can capture signals from unstructured data and combine them with structured data sources, illustrating its potential in aiding businesses beyond just hype.

💡Data Science

Data Science is an interdisciplinary field that uses scientific methods, processes, and algorithms to extract knowledge and insights from data. The script highlights the importance of using generative AI in conjunction with established data science practices to serve valuable business cases.

💡Data Engineering

Data Engineering involves the collection, storage, and processing of data to ensure its quality and accessibility for analysis. The video script mentions that generative AI is best used in combination with data engineering to address business goals effectively.

💡Open AI

Open AI is a research laboratory that aims to develop artificial general intelligence (AGI) in a way that benefits humanity. The script discusses a dramatic event involving Open AI's CEO dismissal and subsequent return, indicating the volatility and rapid changes in the AI industry.

💡Microsoft

Microsoft is a leading technology company that has heavily invested in AI, particularly through its alliance with Open AI. The script mentions Microsoft's involvement in the AI space, including its efforts to preserve Open AI and its strategic partnerships with other AI companies.

💡AI Safety

AI Safety refers to the practices and measures taken to ensure that AI development and deployment do not pose risks to humans or society. The script touches on the challenge of balancing AI safety with commercialization, indicating the importance of this aspect in the development of AI technologies.

💡Cloud Platform Vendors

Cloud Platform Vendors are companies that provide cloud computing services, such as Amazon Web Services (AWS), Google Cloud, and Microsoft Azure. The script positions these vendors as key players in the AI competition, emphasizing their role in driving market share and growth in the AI space.

💡Chip Makers

Chip Makers are companies that design and manufacture semiconductor chips, which are crucial for powering AI models and cloud platforms. The script specifically mentions Nvidia, highlighting its significant market share and influence in providing processing power for AI training and usage.

💡Hugging Face

Hugging Face is a company that provides a platform for developers to access and integrate the latest AI models. The script describes Hugging Face as a valuable resource for developers, helping them understand and utilize the rapidly evolving options in AI models.

💡Forrester Wave

The Forrester Wave is a series of reports that evaluate and rank the top vendors in a given market based on their current offerings and strategies. The script uses the Forrester Wave for AI Foundation Models for Language to illustrate the competitive landscape of AI companies and their market presence.

Highlights

AI demystification efforts aimed at discerning the true potential of technology amidst the hype.

Generative AI's capability to capture signals from unstructured data sources and combine with structured data.

The importance of using generative AI in conjunction with established data science and engineering for business cases.

Drama around Open AI's CEO dismissal and subsequent developments with Microsoft.

Microsoft's strategic moves in AI, including deals with Mistral AI and Inflection AI.

Concerns about the stability and management of Open AI amidst significant investment in the AI space.

The balance between AI safety and commercialization as a critical challenge.

Competition in the AI space among cloud platform vendors like Google, AWS, and Microsoft.

The role of chip makers like Nvidia and AMD in powering AI training and usage.

Hugging Face and Llama Index as valuable resources for developers integrating the latest AI models.

Analyst perspectives on AI foundation models for language from Forrester Wave.

Massive investments by tech giants like Meta, AWS, Google, and Microsoft in AI.

U.S. government spending and bipartisan Senate AI working group's call for increased AI innovation funding.

Improvement of AI models in aspects such as longer input context windows, reasoning, and multimodal input handling.

Economic implications and the potential long-term impact of AI on productivity and the world economy.

The integration of generative AI with other data science techniques and technologies for enhanced decision-making.

Realism in the investment community regarding the macroeconomic impact of generative AI.

Generative AI as part of a larger technological mosaic including sensors, 5G, robotics, and biotechnology.

The necessity for businesses to focus on data and its effective use in combination with AI models.

Transcripts

[Music]

good morning everybody back in October

October the 19th was the last thorough

good data and AI update in

London uh at that

time there was an awful lot of awareness

uh you awareness hype some would say

about AI chat GPT had been launched on

the world at the end of November 2022

and and all of the months up to October

in in in 2023 were really about that AI

hype and

our aim really going into that day was

to demystify

AI uh to help really we thought what we

what we would be most useful trying to

do was to help people discern between

the true potential of the technology and

some of the hype that was building

around it um and so our our aim was to

demystify

uh basic points were that it's not magic

and that it didn't just come out of

nowhere so like all technology it's

built on waves and waves of things that

go before it but during the sessions of

that day we were trying to illustrate in

fact I think we're pretty successful in

illustrating that generative AI is able

to do some things for us so generative

AI is able to help us capture signals

from unstructured data sources and to

comp combine those with the data sources

the structured data sources we're so

familiar with with

using um but also we were trying to

illustrate that generative AI is best

used in combination with established

data science data Engineering in the

service of valuable business cases not

uh you as opposed to just trying to look

for things that generative AI on its own

can do

the side of you know capturing signals

from unstructured data in in many ways

what we were saying is data is or

everything is is is data or everything

can potentially be

data and on the other side we're saying

that generative AI adds to the tools

that we have at our disposal to serve

business cases doesn't replace

everything so generative AI adds rather

than replaces

within a month in fact almost exactly a

month after the event there was some

sudden drama around open AI um open AI

obviously having been at the center of

of things as the the creators of chat

GPT um and in fact on the 17th Friday

the 17th of November uh for some reason

the board decided to dismiss the the the

the CEO Sam

mman reported here in the the Ft on the

18th but you a pretty busy weekend

ensued lots of uh rumors and things

coming out but by the end of that

weekend it seemed like uh Sam Alman

would join Microsoft and take a job

there to develop AI there and that

perhaps 95% up to 95% of the staff

working for open AI would go to

Microsoft and within a day or two of

that well it was announced that Sam

Alman would in fact return as CEO to

open aai and that the board at open AI

would be a different board different

composition so uh a weekend a long

weekend of drama um for one of the

players at the sort of center of of

things now of course open AI had been

you know uh supported heavily invested

in by Microsoft and Microsoft had placed

quite a lot of bets on open AI in terms

of its own strategy so during that

weekend its involvement in trying to

make sure that there was something

preserved out of the bsed place was was

important but also uh by here late

February uh you could see again another

ft clip Microsoft strikes deal with

mistel so mistl AI French AI company uh

had at that point been established

perhaps a year uh gone from startup to

well today valued at something like $5

billion so I know deal with Microsoft

and that really showing that Microsoft

was pushing Beyond its depend or

Alliance that to that point with just

open

Ai and it's believed that you microsof

started that conversation with Mr all in

December so pretty much immediately

after that that sort of blow up at open

AI uh

March Microsoft hes somebody uh the

chief executive of inflection AI but

also the guy who really had been behind

deep mind that was bought by Google uh

mustapa

sullon they didn't acquire inflection AI

that doesn't seem to be Microsoft's

style in this but but somehow Microsoft

is building connection with a lot of

players and particularly a lot of

innovators and startups in in the space

um but you coming much more up to dat by

the end of May still concerns about the

stability of the the sort of structures

that that that owner or manage and

control uh open Ai and in some ways we'd

say that you that concern about

stability there's clearly a lot of money

flowing into to the a very hot

technology space the main players in it

are

very recently established uh so that's a

recipe for for volatility but of course

there's a a very important uh challenge

with AI about safety versus

commercialization or safety with

commercialization so how to play that is

is clearly an important and a pretty

difficult balance to get right so

there's some

volatility it's all almost everything

that I've said so far has has sort of

been about open Ai and Microsoft and in

some ways the hype of 2023 was Ed by

Microsoft and open AI but it's a very

very competitive space and there are a

lot of players many of them in fact

Google as an example would have been

clearly perceived as the leader before

that that sort of splurge by Microsoft

in in in in November

2022 so the cloud platform vendors

Google AWS Microsoft for them they're

competing for market share of the uh

Cloud Market but there you know

obviously with that goes competing for

growth rate of growth to gain market

share in the space and the promise of AI

for them is just too important a source

of growth for them not to compete very

very heavily for so there's you know the

the cloud vendors are key players in

this AI

competition but the model developers are

very important too so in this sort of

second box on on the screen here uh I've

mentioned mistol mentioned open a a lot

but there are many other significant

players meta massive company that has

developed its llama models and put them

into the open source

Community um Elon Musk with X aai active

in the

space got a smile just mentioning Elon

Musk I I I won't go any further um and

and I I'll I'll say a few more things

about that in a minute but um the chip

makers are important so it's very

unusual for us at a Thor good type of

event to to worry about the hardware too

much but the chip makers well

Nvidia you know I can remember the days

of trying to select the right cards for

gaming not that I gamed much but members

of my family did um and now you Nvidia

the the world's most valuable company as

as of a few days ago uh three trillion

dollar valuation market

capitalization uh but at the moment uh

supplying around 85% of the processing

power that is being used by the model

developers and the cloud platform

providers to power AI training and AI

usage uh so an incredibly powerful

position in in the market because of

some of the very distinctive properties

of the processes they make the massively

parallel processing capabilities but not

only that they've really um developed a

software platform that allows model

developers to harness those chips in a

way that nobody there's no other

software platform that is their

equivalent that has the acceptance among

model developers but it also has some

networking technology to connect those

processors together so Nvidia has a very

powerful

position AMD another uh chip designer in

fact uh Nvidia doesn't fabricate its

chips it designs its chips and they're

fabricated by Taiwan silicon uh

semiconductor Manufacturing Corporation

AMD similarly designs chips and the

fabrication is also done by

tsmc uh but AMD now with chips that in

performance terms can challenge and

surpass nvidias is trying to mount a a

challenge Intel struggling with some of

the the the the very um fine grain

processor technology but nevertheless a

significant place and interestingly

Google AWS and Microsoft all entering

this chip uh Market

too also important to our probably much

less talked about but important to

developers are the places that

developers can go to pick up the latest

models when they're

integrating hugging face American French

company uh llama index providing

something that is incredibly valuable to

developers at a time when model options

are coming up fast and furiously and

there's so much to choose from its way

of understanding what's available what's

trending how to use it and how to

integrate um and the final box managed

analytics

platforms uh I've got two here snowflake

data

bricks essentially similar in some ways

both essentially providing

platforms that will run on any of the

clouds or in a multicloud sense across

clouds so

insulating the application layer from

the particular cloud provider Choice uh

to a large degree but doing it in quite

different way so snowflakes approach to

it to come really from a a relational uh

uh

you well making the appeal to people who

have used relational technology to to do

relational uh data warehousing and

making that fast and easy in in the

cloud and in a multicloud way data

bricks coming at it in a more I would

say Innovative way uh harnessing spark

and getting spark technology as a

foundation enabling data engineering

data science and now ai to be done again

AC cross multic clouds but without the

constraint of relational uh technology

or yeah with the option of relational

technology but without that being a

constraint I'll go to an analyst uh

perspective on some of those

foundational models so this is the

Forester wave a AI F Foundation models

for language now the first thing I'll

say there's a lot of uh I suppose

skepticism or cynicism about some of the

analysts work in any of the fields but

in in this field at the moment it's

quite interesting to try and even you

describe who's got leadership and and

and and who's got strategic Advantage

but the what I'll do is I'll do a zoomed

in version in the middle so that we can

look at some of the players in the space

but I'll describe the the overall um

setup of the analysis first of all so

Market presence the size of circle is

indic indicating in Foresters terms the

the the market

presence um if the circle is gray or the

Little Dot is empty has a space in the

middle of it uh that denot and you'll

see that more clearly when we zoom in

that denotes whether somebody has

participated in the Forester study or

not uh so here anthropic is gray haven't

explicitly participated open AI has

participated so that's an important

distinction I think uh the

axes

um going on the vertical axis from at

the bottom weaker current offering to

Stronger current offering now this is a

very fast moving Market it's very hard

to I think classify what's weak and

what's strong in in a pure

sense uh and then along the horizontal

axis weaker strategy stronger

strategy uh again very hard to I think

decide you know well I'll zoom in a

little bit um so in the leadership

positioning position

Google clearly a powerful strategy

clearly some pretty established presence

in this kind of

Technology at the other end of the

spectrum though mistol classified here

as just coming into it well mistra can

you judge it in isolation how you judge

it because mistal in combination with

Microsoft is quite a different

proposition to mistol a

startup um and open AI in many ways you

know showing a lot of Market presence

Microsoft not oh and there's an example

of a hollow dot so Microsoft not a

participant with Forester in its own

right but clearly

presence in the market through open AI

so it's a bit of an odd analysis but it

serves to put the names in the frame for

us that this is a wide open space

there's a lot of competition a lot of

innovation you can't just say somebody's

got this because they're big or somebody

hasn't got it because they're small you

could come from nowhere and indeed

M you know has done and is

doing I'll move on from that uh to

something about investment so you know

2023 there was clearly investment going

in uh a lot of

talk you amongst the you Market

participants private Equity investors

Etc about all of the money going

in but at the moment you know this year

it's expected that about 200 billion

will be invested

by those four players meta AWS Google

and

Microsoft meta by the way whose

technology llama

or whose Foundation model technology

llama put into open the open source

space not even figuring on the previous

slide we don't know why but it's an odd

one to be missing from from that Forest

analysis but colossal amounts of money

being spent by those companies now

they're private companies they're making

Investments those investments will be

depreciated and that depreciation will

show up in their income statements at

some point their p&l and and they will

be you working very hard to make sure

they're showing a profit on on those

charges not a a loss on them so there's

something very real world focused about

the Investments that those companies are

making colossal though they

are in a different sense this is looking

backwards so it's estimated that the US

government uh spent more than 4 billion

last year on procurement for uh defense

related AI just procurement for defense

related so $44 billion last

year a more forward-looking thing uh a

senate a bipartisan which is quite

interesting you don't hear much that's

positive and bipartisan about us

politics at the moment but a bipartisan

Senate AI working group published a

report in May that calls for at least 32

billion per year non-defense AI

Innovation spending 32 billion per year

and the target is for that to be uh

raised and spent starting from 2026 so a

pretty aggressive thing I think it would

be safe to assume that they won't stop

spending on a defense front as well so

an awful lot of money coming into this

space not just from the private players

but from government

players and that's supplemented by

public announcements coming along quite

frequently at the moment from other

countries

uh making in in many cases you know

multi hundred million commitments to

investment in the space I've listed a

few here um but you clearly somebody

like China is going to be pretty active

too although it's not obvious you what

this what the equivalent expenditure

is um so so it's a very active space

some people need to show a return on

that investment government's probably

different to that they able to act on a

a different belief basis they certainly

don't want to be left out of something

that has the potential that that AI

has um so all of this energy going into

those companies producing the models

well the models are being improved very

quickly and there I've just reduced it

to three different facets here but there

are three important facets so longer

input context windows improved reasoning

and the third one ability to handle text

and image together as input prompts and

more generally the ability to handle

text image audio video coming in

multimodel becoming ever more important

in these models but to the first point

longer input context

Windows

um I suppose if I confess my own naivity

year and a half ago I was thinking a

prompt for chat GPT is not that

dissimilar to you typing a a query into

Google obviously a bit more to it than

that because you could continue to

develop context as you as you prompted

but nevertheless I'm thinking you

relatively small things whereas uh some

of these figures here if we take the

table here uh check uh GPT or GPT 3.4

five turbo the input Windows there were

supporting 16,000 tokens converted into

rough pages of text equivalent that's

about 24 pages of text so even

3.5 a enabling a lot of prompting to be

to be put

forth

128,000 take takes that up to probably

you know getting on for a hundred pages

of

text um by the time you get to yeah a

million tokens

uh I won't do the arithmetic for you but

but many many pages uh that that we can

put in to to these things so you think

well this isn't like T typing a query

into Google this is something more

significant that than that

now in a recent Thing by uh one of the

sort of uh bloggers that I trust in the

space Andrew in he was talking about

multipay prompts as being really Mega

prompts well you'll see later this

morning in this session actually when we

get to the Oxford university endowment

management application the prompts that

we're providing there are many many

pages long and and quite sophisticated

in their structure so uh the ability to

really create very intelligent prompts

for these models is is or very big

prompts I should say is is one important

facet the next Point improved reasoning

Builds on that because the more

reasoning power that is being developed

into the foundation models the better

they're able to interpret and make sense

of large prompts so it the the two

things feed themselves bigger prompt

Windows better ability to make sense of

the prompts and then the ability to

handle text and image well you we might

not have ever thought of putting a

diagram into Google but um you certainly

can start thinking now about putting

images as well as text into your

prompting now there's a there's a cost

Associated so yeah I've put here the the

pricing first of all for the open AI uh

3.5 40 and um yeah four

turbo price of four turbo $10 per

million tokens versus 50 cents per

million tokens so it gives gives you

something more but you pay a lot lot

more for it so these guys are trying to

commercialize this space it's not it's

not a

charity

um table somewhat inconsistent I put

anthropic Claude 3 Opus there that's the

most powerful model at the moment coming

from

anthropic much more expensive relatively

speaking but it is their very top

offering they have cheaper offerings too

uh or yeah good Google Gemini 1.5 Pro

that you colossal uh prompt window

capacity priced relatively competitively

and the pricing here I've put is for um

prompts that are over

128,000 tokens but if you can keep your

prompts within a 100 you keep within 100

Pages you could pay half that price

three $350 for yeah per per million

tokens so pricing no doubt will move and

move and move competitively and quickly

but generally people will be trying to

get more for the latest better

capability um in the last one of these

we kind of built a a pretty elaborate

picture of what's happened in terms of

waves and layers of Technology since the

1970s right up to the current with that

sort of Starburst moment at the top of

the diagram around Google releasing its

file system and and reduce papers which

started something well what did it start

that that Google release started the

cloud era the Big Data era and on top of

that the AI era has been built uh in

some ways you know eclipsing what has

happened over decades with relational

and before that multi-dimensional

technology all of those relevant to us

though yeah as a business thorough Goods

business is helping customers make

better faster decisions

with the data that they have available

so all of these layers have been uh

applicable and still are in in various

ways uh applicable the technology just

gets better the options get richer and

that continues uh with some of the data

handling capabilities graph databases

geospatial databases you'll see examples

of the applicability of those later this

morning in some of the cases but and

Vector databases you'll see uh reference

to that in the Oxford university

endowment management case so these

Technologies just layering in at the

cloud big data and AI level enriching

what we can do so it's not all about

large language models or Foundation

models it's about the combination of

those models with other data science

techniques other data engineering

techniques and and platting it together

effectively um

um greater realism I think one of the

big differences between you know

standing talking today and even eight

months ago there's much greater realism

if I talk to people in the investment

community at the moment they've realized

that it's not just about large language

models it's not just about generative AI

it's about data engineering capabilities

data science capabilities business focus

yeah what you do with it um but there

was a a point the world economic Forum

um some economists saying stay stay

realistic about generative ai's

macroeconomic impact but it always

amuses me when economists tell us to

stay realistic but but um putting that

aside and looking at one of my

colleagues who might be upset by that

but

um you know stay realistic about

generative ai's macroeconomic impact

I the economists making that statement

weren't saying that there won't be much

impact but they're saying it's likely to

take

longer than instant it won't be

instantaneous that the effect is felt

but even small impacts over long periods

of productivity difference will be

enormously beneficial to the world

economy so it's too important to ignore

the first bullet point though generative

AI is is a critical piece of a

technology technological

Mosaic including sensors 5G robotics

biotechnology etc etc is part of a

mosaic and the second bullet point

exuberance typically accompanies

remarkable

Innovations but I think when an

economist uses

exuberance they're probably referencing

back to the irrational exuberance phrase

that Alan Greenspan used to use when he

was uh chairman of the Federal Reserve

in the United States and he was using

that to describe what was going on in

the do com boom of the 1990s and into

into the uh early 2000s so the doom boom

people thought that was a boom that

burst and bust but out of it came Google

Amazon I mean that wasn't irrational but

there was a lot of stuff around the ones

that did survive and make it big that

was much more irrational than that so

we're in a similar phase I suspect where

it's sensible to be grounded and the

theme that we'll be pursuing today is

how to ground that how to ground in

things that will be valuable for your

business

um so coming back and and in some ways

relating to that that that previous

Economist Point um generative AI as I

said at the beginning is best used in

combination with established data

science and data engineering ing

capabilities to address valuable

business goals rather than looking for

things that generative AI alone can

do generative AI adds rather than

replaces we'd say in fact data is

everything data is the basis for

training these foundational

models that's what's enabling them to

develop their intelligence and their

usefulness but

the data that any particular company has

uniquely to itself the latest

information particularly the most unique

information that a company has that can

be used in combination with those models

to apply the strength of those models to

the strength of the company's contact

with its customers in the world so it's

about

using data effectively data is

everything and as we said earlier

everything is dat as well in this world

where large language models can look at

unstructured information and pull

signals out of it so everything is data

data is everything it's not magic it

didn't come out of nowhere and I'll hand

over to Amanda

[Music]