Freeing up newsroom time: How Amedia’s AI sandbox will allow reporters to do more

Summary

The video is abnormal, and we are working hard to fix it.
Please replace the link and try again.

Takeaways

• 🌐 欢迎参加第三届创新本地网络研讨会，有来自30多个国家的人士参与。
• 👩‍💼 Cecilia Campbell H 是创新本地项目的编辑，与 Nicholas Yason H 一起推动该项目。
• 🤖 本次研讨会的主题是挪威地方媒体集团推出的全新AI沙箱，旨在为记者提供工具。
• 📰 创新本地是首个全球本地媒体论坛，旨在促进媒体专业人士交流想法和解决问题。
• 🌍 研讨会通过Zoom平台进行，鼓励参与者在问答环节提问。
• 👨‍💼 Marcus Rasanen 是挪威地方媒体集团新闻部的主任，分享了他们如何利用AI沙箱。
• 🚀 AI沙箱的目的是帮助记者节省时间，提高工作效率，同时保持内容的真实性和可信度。
• 📈 媒体行业面临全球性挑战，如成本上升和通货膨胀，AI技术可以帮助应对资源减少的问题。
• 📝 AI沙箱不仅是工具，还是一个实验平台，可以让记者在安全环境中尝试不同的AI功能。
• 📱 通过AI沙箱，记者可以快速生成文章摘要、改进标题和寻找采访源。
• 🔍 强调了在使用AI工具时，记者需要进行事实核查，确保内容的准确性。
• 📅 下一次创新本地网络研讨会将在3月20日举行，主题是受众发展和社区服务。

Q & A

创新本地网络研讨会的目的是什么？

创新本地网络研讨会的目的是为本地媒体专业人士创建一个全球论坛，以交流可行且可扩展的想法，特别关注本地新闻业务所面临的机会和问题。

Cecilia Campbell H和Nicholas Yason H是如何参与创新本地项目的？

Cecilia Campbell H是创新本地项目的节目编辑，而Nicholas Yason H是节目主任。他们两人将创新本地的理念带到了WAN-IFRA，并与他们合作交付该项目，汇集了来自世界各地的本地媒体专业人士。

Marcus Rasanen在创新本地网络研讨会中扮演什么角色？

Marcus Rasanen在创新本地网络研讨会中扮演评委的角色。

Outlines

00:00

🌟 创新本地网络研讨会介绍

本次网络研讨会是创新本地项目的第三次定期活动，由Cecilia Campbell H主持，她是创新本地项目的编辑，与她一同出席的是项目总监Nicholas Yason H。他们介绍了创新本地项目的宗旨，即创建一个全球性的本地媒体专业人士交流平台，分享可操作和可扩展的想法，专注于本地新闻业务的特殊机遇和问题。此外，他们还邀请了挪威本地媒体集团的新闻总监Marcus Rasanen，分享他们最近为旗下千名记者推出的全新AI沙箱，并讨论了他们希望通过这个项目实现的目标。

05:01

🤖 AI沙箱的潜力与挑战

Marcus Rasanen分享了关于AI沙箱的见解，他认为即使小出版社或独立报纸新闻编辑室也可以利用现有工具完成80%到90%的工作，而无需太多开发工作。他谈到了自己在新闻业的背景，以及A媒体在挪威的业务模式和挑战。他强调，尽管AI技术提供了许多机会，但信任是新闻行业最宝贵的资产之一，他们希望通过AI沙箱帮助记者更有效地与人们交流，而不是简单地生成更多相同的内容。他还提到了A媒体如何通过制定指导方针来负责任地利用AI，以及如何保护内容不被用于训练大型语言模型。

10:02

📰 利用AI沙箱提高新闻工作效率

Marcus Rasanen通过实际演示，展示了AI沙箱如何帮助记者提高工作效率。他上传了一份关于市政会议的PDF文件，并使用AI沙箱对其进行了摘要，快速提取了关键信息，为撰写新闻稿件提供了基础。他还展示了如何使用AI沙箱生成新闻标题，并强调了在使用这些工具时需要精确的提示。此外，他还提到了AI沙箱的其他功能，如将文本转换为音频，并强调了在发布前进行事实核查的重要性。

15:05

🔍 AI技术在新闻编辑室的应用

Marcus Rasanen讨论了AI技术在新闻编辑室中的应用，包括如何使用AI工具来提高记者的工作效率和创造力。他提到了A媒体已经实施的一些项目，如使用AI分析编辑内容、个性化服务、音频转文字等。他还介绍了A媒体如何通过AI沙箱收集使用数据，以便改进和开发新的工具，如自动写作者和事实核查器。此外，他还强调了在安全环境中实验AI工具的重要性，以及如何通过这些工具传播专业知识和技能。

20:06

📢 与观众互动和未来计划

在网络研讨会的最后阶段，Cecilia和Nicholas回答了与会者的问题，并讨论了AI在新闻制作中的透明度问题。他们强调了向读者明确告知AI参与新闻制作的重要性，并分享了A媒体如何通过AI沙箱收集反馈和改进工具。此外，他们还提到了未来的计划，包括将AI沙箱整合到内容管理系统中，并根据记者的反馈继续开发新的功能。最后，他们邀请与会者参加未来的网络研讨会，并鼓励大家在公司内部传播创新本地项目的信息。

Keywords

💡innovate local

创新本地化（innovate local）是本次网络研讨会的主题，旨在为本地媒体专业人士提供一个交流可执行和可扩展想法的全球论坛。它强调本地新闻业务特有的机遇和问题，尤其是关注本地新闻市场的挑战。

💡AI sandbox

AI沙盒（AI sandbox）是指挪威地方媒体集团为旗下记者推出的一个全新的人工智能实验平台。该平台允许记者在一个安全的环境中尝试和使用各种AI工具，以提高工作效率和内容创作质量。

💡local media

本地媒体（local media）指的是在特定地理区域内运营的新闻媒体公司，它们通常关注当地的新闻事件和社区动态。这些媒体机构面临的挑战和机遇与国际或全国性媒体有所不同。

💡digitalization

数字化（digitalization）是指将传统的纸质媒体内容转换为数字格式，并利用互联网和移动设备等数字技术进行传播的过程。这一转型对于新闻媒体来说至关重要，因为它有助于扩大受众范围并提高运营效率。

💡editorial and business topics

编辑和商业主题（editorial and business topics）涉及新闻内容的创作、编辑、发布以及媒体公司的商业运作和盈利模式等方面。这些主题对于新闻媒体公司来说至关重要，因为它们直接影响到内容的质量和公司的可持续发展。

💡subscriptions

订阅（subscriptions）是指读者为了定期接收新闻媒体提供的内容而支付的费用。对于许多新闻媒体公司来说，订阅收入是其主要的收入来源之一，有助于确保稳定的收入流和读者忠诚度。

💡fact checking

事实核查（fact checking）是指验证新闻报道或其他信息来源中的事实准确性的过程。在新闻行业，事实核查是确保报道真实性和可靠性的重要环节，有助于维护媒体的公信力。

💡trust

信任（trust）在新闻行业中是极为重要的概念，指的是读者对媒体提供的信息真实性和可靠性的信念。建立和维护信任对于媒体机构的长期成功至关重要。

💡local identity

本地身份（local identity）是指新闻媒体与其服务的特定地区或社区之间的紧密联系和认同感。本地媒体通过关注本地事件、文化和居民的需求，来塑造和强化其本地身份。

💡live demo

现场演示（live demo）是指在网络研讨会或现场活动中，演讲者实时展示产品、服务或技术的过程。通过现场演示，观众可以直观地了解产品的功能和操作方法。

Highlights

本次网络研讨会是第三届innovate local定期网络研讨会，吸引了来自全球20-30个国家的参与者。

innovate local是全球首个专注于地方媒体的论坛，旨在促进媒体专业人士交流可行且可扩展的想法。

innovate local项目由Cecilia Campbell H和Nicholas Yason H共同发起，他们都有国际媒体行业的工作经验。

特邀嘉宾Marcus Rasanen是挪威地方媒体集团新闻部的主任，代表了创始会员公司之一。

Marcus Rasanen将分享他们最近推出的AI沙箱，以及他们希望通过这个工具实现的目标。

innovate local项目旨在为新闻媒体公司提供国际论坛，关注地方新闻业特有的问题和机遇。

网络研讨会每月举办两次，每次由一位出版商案例呈现，并聚焦于他们如何解决一个明确定义的问题或机遇。

AI沙箱的目的是为记者提供一个工具，帮助他们在本地新闻室中节省时间。

AMedia拥有超过120家报纸和超过1000名记者，覆盖挪威全境。

AMedia的商业模式主要面向订阅用户，拥有超过700,000名订阅者和超过2,000,000名每日读者。

AI技术的目标不是取代记者与当地社区的互动，而是支持和扩展这种联系。

AI沙箱的推出是为了帮助记者更有效地利用与人们的交流，提取更多信息。

AMedia建立了一些关于如何负责任地使用AI的指导方针，以帮助编辑确定他们如何接近AI。

AI沙箱是一个安全的环境，允许记者自由实验AI工具，而不必担心数据安全问题。

AMedia的AI沙箱是一个动态发展的平台，它根据记者的使用情况不断更新和改进。

AMedia通过AI沙箱收集的数据来指导产品开发，使记者的使用模式直接影响新工具的开发。

AMedia计划将AI沙箱集成到记者的工作流程中，使其成为日常工作的一部分。

innovate local网络研讨会提供了一个平台，让地方媒体专业人士分享和学习如何利用AI技术提高工作效率。

Transcripts

all right let's get this show on the

road so hello and a very warm welcome to

our third regular innovate local webinar

it's fantastic to have people joining us

from over um 20 30 countries across all

continents uh and I do hope you'll find

today's case valuable and inspiring um

my name is Cecilia Campbell H and I'm

the innovate local program editor and

with me I have program director Nicholas

yason H and the two of us took the idea

of innovate local to WAN ifra uh and

we're very happy to be partnering with

them to deliver the program bringing

together local media professionals

across the world um Nicholas and I have

both worked in the media industry

internationally uh but we're now back in

our country of origin which is Sweden

Nicholas lives in Stockholm I live in

malma in the South and today we're

really happy to be joined by Marcus

rasanen who um is director of news at

Norwegian local media group or media

we're extra happy to have Marcus here as

he represents one of our founding member

companies so thank you for your support

Marcus it's much

appreciated Marcus is going to talk

about yeah good to see you Marcus is

going to talk about uh the brand new AI

sandbox they launched to their thousand

thousand journalists last week and what

they're H hoping to achieve with that if

you have any questions during the

webinar please put them in the Q&A

section at the bottom of the zoom

interface and then we'll have a Q&A uh

uh session at the end of the

presentation before I hand over to

Marcus let me just say a few words about

this

program so innovate local is the first

global forum for local media the idea is

to create a space where media

professionals can exchange actionable

and scalable ideas with a focus on

opportunities and problems that are

particular uh to the business of local

journalism um the program is aimed at

anyone working in news media company

which operates in its own geographical

community so until now there's been no

International Forum with a focus on the

particular problems and opportunities of

local media so uh yes newsrooms and

markets defer from region to region from

country to Country so I think that's why

there's not been anything till now but

we've see is that the needs of local me

people and local businesses have more in

common than not across different

countries so that's our starting point

there's no Silver Bullet but there is

lots of ground baking work being done by

Folks at local news Publishers all over

the world so we want to bring you all

together and this program is for you and

by you we're delivering two case

webinars a month presented by one of you

uh each webinar features one publisher

case and focuses on how they worked with

one clearly defined problem or

opportunity so the webinars take place

every other Wednesday at 3:00 European

time uh and they will cover all kinds of

editorial and business topics which are

particularly relevant to local media

some will be more of more interest to

you than others depending on what your

role is at your company so the idea is

that you can tap into this resource when

it suits you and of course all webinars

will be available to watch On Demand

afterwards so we'll publish the webinars

write up of cases and other resources

all on the innovate Local website so

have a look there all right let's get on

to today's case Marcus welcome to

innovate locco thank you so much for

being here today to tell us about how AR

media is hoping to free up time in the

local newsrooms by creating an AI

sandbox as a tool for all your local

journalists I'll let you explain

how all right yeah I'll try thank you

very much for the introduction Cecilia

I'm very very very happy to be talking

to all of you so the most nerve-wracking

moment of any webinar of course is

sharing your screen no amount of AI

sandboxing will will help us do

that uh I think we're good can you

confirm

Cecilia I can conf I can confirm yes

perfect

yeah so um yeah um I'm really thrilled

to be part of uh of this webinar series

because I think it's um it's something

that the international scene for Journal

journalism has lacked as cilia so

eloquently explained um for us um we

wanted to sort of um put the spotlight

on the opportunities that AI represent

and also uh try to make it as relevant

as possible for even smaller Publishers

or Standalone newspapers newsrooms uh

it's my opinion that what we do here uh

you can probably do 80 to 90% of it uh

with little to know development work

just by using available tools today so

that that is the basis of what I'm going

to show you I'm going to talk a bit

about our Challenge and our setup and

our IDE of soling that and I'm also

going to do a live demo uh hopefully

it's going to go as planned but this

things are uh notoriously unpredictable

so let's

see um just a couple of words uh about

myself and my background I'm the

director of news at am media uh I worked

in two local newspapers in Northern

Norway for 11 years I held almost every

role imaginable in those uh those titles

uh and I've also always been interested

in technology and de

development uh and actually at one point

in my life I decided while working in a

newspaper to found a company that um

developed and sold publishing systems to

small newspapers which was ironically

bought by am media before I started

working for am media so that was that

was a fun twist uh since 2023 I've

worked in our Central organization so

not in any newsrooms but um but as um as

part of the central organization

here so yeah uh first of all it's

important for me to um most of you are

probably somewhat familiar with what a

media are but for those who

aren't uh we consist of more than 120

newspapers all over Norway which is

quite a long country as you might have

noticed so basically we have more than a

thousand journalists uh situated all

around the country uh we are more than

700,000 subscribers our uh business

model is primarily uh geared towards

subscriptions wasn't that way before but

we've done a huge am amount of work in

terms of digitalizing our subscriber

base and also we've got more than 2

million daily readers so that's that's

who we are uh if you can see my pointer

now I'm trying to point out where I

started my career in local news media in

a town called

Buddha I noticed a couple of Swedish

names in the chat uh so you probably

heard about the local football team

there Buddha glimp which I'm quite proud

of just needed to get that in there okay

so um our local newspapers and newsrooms

are our core that's the core of a media

that will never change so everything we

do is uh is geared towards solving the

problems that arise locally that's a

really important principle for

us so when we're approaching all these

questions all these uh possibilities all

these challenges that artificial techn

technology and artif artificial

intelligence presents it's very

important for us to remember that we

need to to do it in a way that supports

uh our local news rooms we're not trying

to uh the aim for us is not trying to uh

you know just uh work more effectively

on a news desk providing even more of

the same content uh that uh pretty much

anyone can do generically our goal is to

support um the meeting between humans

our presence in our local societies and

uh and our local identity so that's just

something I wanted to to state before

the

presentation okay so this is a picture I

I like uh very much it's O yanes to the

left here he's a journalist in a small

newspaper that we own in the west of

Norway that's called

Fon his uh on this picture his um

recording a podcast Series where he

interviews a man to the writer called

BNA who's dying he's got a cancer

diagnosis uh B actually died a couple of

weeks ago and olanes has made a quite

touching podcast Series where he

interviews B about his uh perspective

towards life itself I might say um it's

it's a really touching touching series

if any of you speak any Norwegian I

would recommend it but the point about

this picture is I think it's uh you know

it's just uh it's just made his own

podcast studio uh in a cabin and this is

basically what a media is and the AI

technology can never replace this you

can never replace meeting humans uh and

getting them to tell their

stories so that's that's sort of our um

our objective here it's not to replace

this it's to expand upon

this uh as you all know uh we've got the

global challenge going at the moment in

terms of uh Rising costs uh inflation

and in Norway that's quite

brutal uh most media most Publishers and

most

newsrooms uh are going to have to make

to with less people than they did just a

year or two ago because of the global

economy and the

and the things arising the problems

arising from that so that's a problem we

need to solve we need to serve our

readers and our users with less

resources than we did before and they

kind of expect more in return because uh

they measure us against Global uh

entertainment providers or other

providers of um of

content and obviously this uh this is

quite the conundrum so the question we

tried to to answer or start answering

with our AI sandbox is how can we help

oanes here to extract even more from

meeting people like

P so that's one part of the equation

we're trying to solve with the AI

sandbox the other part and perhaps one

of the biggest challenges when we as a

media industry face the opportunities

presented by AI is

trust because I'm of um the opinion that

trust is one of the most valuable things

uh that we have

in um in um the global news industry and

obviously this trust as you can see on

these slides uh might be you can say

challenged by the opportunities that AI

provides this is a small newspaper in

Norway that uh probably have discovered

mid Journey or gencraft or image

generator of sorts and uh started

illustrating their articles uh primarily

using that

I would not Advocate that I don't think

that's um I don't think I don't think

that's the way to sort

of um sort of

uh you know we we need to we need to

never lose sight of the trust Dimension

and if you start doing this I think

there's a danger that uh we just uh more

of the same kind of content stream

that's going to pop out uh pop up in um

more frequent variety that we have seen

lately because of AI possibilities and

also you know generating photo realistic

is images I would not recommend doing

that but some local newspapers in Norway

have tried to do that um I'm not really

sure about the results so this is

definitely not what we're trying to

solve with our a

box uh trust as I mentioned is one of

our primary Commodities and to help our

newspapers uh exploit AI in an

Innovative but responsible way we've

established some guidelines uh for them

that will help their the editors uh

determine how they approach AI basically

these are not strict rules every editor

must determine their own rules but I

think they need some guidance and we

tried to provide that for instance we

don't recommend using photo realistic AI

generated images to illustrate our

articles that's just one one example

obviously we don't Advocate uh putting

sensitive information into third party

services and so forth so the trust

Dimension is important for us to sort of

solve when we're trying to do this and

also if you look around yourself in in

the world today there's a bigger picture

uh as I mentioned we're quite heavily

geared towards subscription Revenue uh

we want to protect our content uh from

uh being used to training

big llms large language

models uh if our content is going to be

used to train llms we want to know how

it's going to be used and we want to

control it so that's uh a principle that

we try to apply when when making our own

AI sandbox and basically we want the

freedom to be able to experiment with

editorial use of AI in a safe

environment which you really can't do if

you just uh publish or feed unpublished

articles into chat gp4

um so that's another part of the problem

we tried to to

answer um in Amia we have actually

explored AI for quite a few years uh

since 2018 we we had our own dedicated

team of AI Specialists that supports all

our lines of business um so it's not

entirely new ground uh but we have

focused our efforts towards other things

than um than sort of the products or the

improvements that uh that meet the

readers and users and our journalists

we've we have trained a lot of language

models that helps us analyze and

understand our editorial content we've

done a lot with personalization

obviously we got a huge amount of data

since we have so many subscribers which

we have good and structured data about

and also we used it for simple tools

like converting audio to text and and

such um one of the big projects which we

want to exploit more the coming years is

how we can use our AI plat platform and

our structured content data um and how

that might be able to power up our uh

publishing system because obviously if

you can integrate uh AI tools directly

uh into the workspaces of our

journalists then that's very interesting

and this sandbox is kind of the first

step towards that but not quite I'll

show you the difference in the demo

okay so that's the background and the

reason why we want to do this um I think

this is where I try to go live uh with

my demo so that's another nerve-wracking

moment just a couple of seconds and I'll

try

to

uh uh

okay sorry I need to hide a couple of

panels here um can you see my screen

Cecilia yes I

can yeah perfect it's good okay so this

is basically the interface of our AI

sandbox when uh every one of our more

than a thousand journalists uh go into

this sandbox they they're met with this

interface and sort of the idea here is

twofold for one we want to control the

content that we put into it we don't

want it to be used or abused and the

second part of it which which is kind of

the coolest thing is um we're using what

we call our journalistic hive mind in am

media as I said we're more than a

thousand journalists situated around the

whole country of Norway and what we do

when when these journalists use this

sandbox is we're logging their use and

we're getting IDs for product

improvements Direct IR ly from the use

cases that they prompt into our models

so just to give you give you an example

if um if a large part of the use is um

making U or if um if someone uses this

to summarize articles then obviously we

can make functionality that does exactly

that which we've done here if uh another

use case is uh writing better headlines

then obviously we can Implement that or

develop that capacity and so we want the

organic use of our journalists to be our

guideline when we develop their

workspace basically and I'll show you a

couple of things which we've already

implemented and this is uh this is a

Sandbox that is a living thing I'm

pretty sure there are developers right

now doing something under the hood there

so I'm I'm hoping this uh beta version

is is cooperating with me so first I

just wanted to give you a C of ideas of

how this would this will help our

journalists in their their um their work

uh for instance a typical task for a

local journalist in Norway is keeping

track of all the these uh municipality

meetings or Council meetings or planning

meetings uh I just went and brought a

report from my hometown of Buddha here

which I found on the web page and I'm

going to do this demo in English

obviously because language is is not a

barrier here so here I've just uploaded

the PDF with uh with information about

the meeting that's going to be held and

obviously you can do simple things like

uh prompting that you're a journalist

and you want a summarized version of

this for instance if you you can't be

bothered reading through the

documents

and obviously the better the prompt as

you probably know the more concise the

output

is so let's see if the sbox corporates

here yeah so here I got a couple of

buildings the city of Buddha has receiv

received an application from nor gra

this is which is a travel traveling

agency for co-financing of marketing for

a direct route between Buddha and

Helsinki and I can read that fer has

decided to do a direct route between

Buddha and Helsinki which is obviously

an interesting story for me to write

about is local journalist

Sima

um so I want to uh speed up the work a

bit and obviously you need the prompts

to be precise which is a large part of

implementing the sandbox now um and

helping our journalists use this eff

efficiently of course we use quite a lot

of time uh

training

so let's see what what we come up with

here okay that's that's fine and also I

can read a quote from the mayor which is

fine which I'm quite sure the AI have

invented because budha does have a

different different mayor now so just

just a a case in point when it comes to

fact

checking what you're going to do but

okay uh

could I want the more Tablo headline I

don't think that's that will cut

it okay I like the last one there

and I don't want the quotes from the

mayor okay so now basically in my

mind uh I've got what I might call a

base for an article not a finished

article it's not fact checked I not

spoken to anyone but with a couple of

minutes of work and prompting I

basically uh I done a lot of generic and

uh frankly quite unnecessary work so

using this you can then expand upon this

you could for

example ask the sandbox which sources

you would

use

and I'm pretty sure you get decent

responses what's interesting for those

of you who are watching this and think

oh this this is cool but I can't do this

with my local newspaper well you kind of

can I don't think it's it's really that

difficult uh basically this uses chat

GPT Force engine and we've just built an

interface for it uh what I would say is

I would not recommend uh feeding uh

information unpublished information

sensitive information into chat GPT

directly I would definitely recommend

putting it in a safe environment which

would not be a huge body of work to get

done but basically by just uh just a

couple of easy prompts I've got the base

for an article and I got a list of quite

good uh sources which I could interview

about this so that's just one example uh

of how you can use this as a journalist

to become a bit more efficient and to to

save a bit of time um I wanted to do

another example just

quickly so yeah uh let's imagine that

I'm a journalist still in my hometown of

Buddha that's uh that just started work

uh

and and my editor has told me that

you're going to do a follow-up story on

this story which um which is about

schools that are going to be to be uh uh

to be close because of economical

reasons basically I don't know anything

about the story beforehand I just

started working today so what do I do

well I can uh save a bit of time by

using the AI sandbox to summarize

this and as always I try to

be as concise as

possible

okay looks quite good quite easy to get

a handle on what this is

about

you and as you can see uh still I just

used basically I've only used the chat

function which is to say most of this

you can do using just the chat functions

from uh large language

models uh and as I said it's probably

not a big development work just to get

that in a safe environment uh we also

got some more advanced uh Advanced

functionality here we have the ability

to uh to to tr to get audio from text

here using different voices this is also

not technology we've developed ourselves

it is a provider that we just uh made

available in the user interface but it's

in the same controlled environment so

that's one option we've got and we

obviously got stuff like uh headline

proposals and and and really quite basic

quite basic functional functionality at

the time being but uh if I go back to my

presentation uh there are some

opportunities that open up when our

journalists start using this these tools

and starts getting familiar with the

opportunities that they provide

wide so we've launched an editorial AI

Hub uh which have a couple of initiative

leads that work in our Central

organization the intention of this AI

Hub is to connect basically our

newsrooms uh the journalists out there

which of which eight are participating

in this Hub project with our development

resources and our AI team and our

editorial development team so basically

uh they will try to be ambassadors they

will try to learn how we can use these

tools in a most Innovative way and we'll

try to let that also influence the way

we product developed so this is how it

works we put something into our AI

router uh which uh obviously

communicates with open AR or Google or

any other third party and we get

output we log the data so we know what

our journalists use this

for we can give feedback towards the

output thanks to our AI team and then

basically over time we can build our own

am media GPT which knows am media's

content which knows uh what constitutes

a good headline and a less good headline

which knows what journalists really mean

when they type different prompts into

the interface so that is sort of um sort

of the

ID which we just started down the rout

of

uh but of course even in a safe

environment even in a sandbox there are

dos and don'ts so what we tell our

journalists in addition to giving the

editors some guidance on establishing

their own guidelines we obviously tell

everyone to use the sandbox don't be

afraid to use it you can't do anything

really wrong as long as you don't I

don't know feed your personal health uh

information or something into the tools

um and we need to get them to start

start asking the question can this

sandbox help me with this so uh that's

probably the biggest uh the biggest uh

body of work it's getting enough people

uh interested in and getting them to

understand the possibilities that these

tools give

you uh obviously you need to fact check

before publishing which by the way goes

for uh not just gen AI

created content but pretty much every

content but it's uh it's even more

important now obviously and also there's

no don't there's a couple of don'ts

which uh you can pretty much um if

you're if you're sensible about it you

can pretty much uh deduce this

yourselves but it's it's nice to be

explicit about

it uh and obviously there's an

opportunities opportunity to give

feedback directly to the sandbox as

well um so if I Tred to summarize why

did we make it well uh it gives us

flexibility in terms of which third

party AI tools we want to use in a safe

environment it ensures data

security uh it's cost efficient and we

have control over the costs we don't

distribute the decisions about which AI

tools I want to buy or plug into our

systems uh into 100 news

rooms and of course we gain knowledge of

our business needs and the potential use

cases through the use patterns we see

also this is a chance for our

journalists to yes as I said guide the

direction of our product development

which I think is maybe the most exciting

thing about all this and it's also a way

to ensure that we spread expertise uh

and um the ability to use AI tools

efficiently as a journalist to us an

editor throughout our

organization

and we have had a couple of workshops

with the AI Hub these are just a couple

of things that we want to implement into

the AI sendbox quite uh quite quickly uh

using uh the feedback that they have

given us so for instance the

brainstormer which is basically a more

efficient way of uh asking for IDs fact

checking obviously is going to be

important Auto writer is something that

we're working on at the moment basically

uh for instance if if a journalists uh

in a local newspaper this happens quite

frequently uh in our

newspapers if someone wanted to do a

story about the weather then they might

check uh the weather forecast and they

start write up writing up a story the

auto writer can do this for you you can

call the meteorologists or something and

get get a quote instead and be more

efficient talk to

people um and obviously different um

different kinds of

functionalities this is just a result of

one workshop and the initial feedback

we've got gotten from journalists and

this is going to expand quite quickly um

at the moment this is a Sandbox which is

a stand loone product it's not

incorporated into our CMS but obviously

uh that is the end goal we want to incor

incorporate this this into the workflow

of every journalist but just to

reiterate it you don't need a host of

developers you don't need a huge Central

organization you don't need to plug this

into your CMS what what you can do is

hire a developer and let them spend 10

or 15 hours to set up this safe

environment for you um the main body of

work um with our a andbox has basically

been done in about a month with one or

two people involved so it's not a lot of

work it's not Out Of Reach for even

really small

Publishers and obviously our our wish

here is to create a set of tools that

helps our journalists become better in

their jobs and enables time uh spent

meeting real people and writing real

stories so that's basically the Crux of

it uh I'm afraid I did the the live demo

a bit early in the presentation but I

think it works um pretty much worked

either way don't you think

Cecilia absolutely that was brilliant

very seamless I have to

say um thank you so much Marcus um I'm

going to let we have we got a question

we need to make sure we have enough time

for questions we have a question from

Saka ludik uh she said during the demo

she said it looks like there's a

limitation of use for your reporters the

interface says something with usage of

tokens she's wondering what this is

about yeah it's basically it's basically

functionality that exists in every large

language model because it it's so

resource demanding but you don't see it

in the interface of chat GPT because so

basically what it means is um in every

context window you have a limited amount

of

memory in the language model so for

instance if I were to chat for two more

hours about the story I chatted

about earlier then I would use up sort

of the the context memory of that

particular um instance but what what we

basically do is refresh the window and

start over again

without the earlier context so it's not

a limit and it's very cost efficient it

doesn't cost us a lot at all so that's

that's just the way the large language

models Works uh chat jpt does this but

you don't see it in the interface so

perhaps we should hide it

that's and and uh I think we saw three

different large language models there um

why have you done that and what is the

learning so far well it's only been a

week

but yeah but obviously the the use cases

are quite extensional already so we have

learned a lot already just during the

first week but but basically the idea is

to we don't want to limit uh the access

to different types of language models or

AI tools we want uh to experiment with

which ones perform the best and which

ones uh are the most intuitive for our

journalists and the early findings is

that chat GPT 4 is by far the most

popular and

better one so far

so G is not not that no and they ran

into trouble didn't they yeah yeah yeah

they did and

[Music]

um a question about fact Checkers yeah

sorry how does check fact Checker

works yeah so basically we have um this

is um this is a system deviced from our

AI team the reason why why large

language models hallucinate is because

they predict uh but they can only

predict uh based on the context that you

give them so basically if you if you

Anchor

Information then you can use llm to fact

check for you I don't remember the

technical term it's a different type of

model but basically you instead of uh

asking the training material of chat GPT

who's the president of the United States

you anchor uh a verified place of

information for instance Wikipedia if

you trust that or another encyclopedia

and then you go there

directly that's one way of doing it

right uh I have I have a quite a

different question but I'm quite

interested in knowing are you

communicating this to and do you think

it's necessary because it's not actually

you're not autop publishing anything but

are you communicating to your readers

about the fact that AI is involved in

your journalism now somehow and how do

you do that absolutely so this is um

basically the way am media works is this

is every editor's uh own privilege to to

to decide how you want to do this but we

we've established some guidelines as I

showed you and in those guidelines we

for instance recommend uh that whenever

you've used AI tools you should probably

explicitly tell the reader about it so

for instance this tic this article

summary has been produced with the help

of AI tools or right a text yes that

sort yeah absolutely but definitely it's

it's that's a really important part of

the the trust Dimension I think to be

open where

and um we have a question from um Dean

uh every editor and journalist has exper

access to this right there is no limit

who can use it no that's

right

[Music]

um

the kind of the problem you know every

editor and has access to chat GPT as

well and we don't want them to use them

use it without

thinking yeah but they don't have to use

it yeah do they have to use it um

interesting question I know where very

good question uh I know shipstad for

instance some of their newspapers have

implemented this it as a demand that you

have to use AI tools just to get used we

have not done that yet we won't to

inspire people to use

it but I'm I don't think you can

really you you you you do not have the

choice of not U trying out some of these

possibilities as a journalist I don't

think you can I just don't think it's

viable no I think you're right in this

day and age

nichas you have any

questions yeah so you had it out for

some for a short time it's quite kind of

new right so so what are the First

Reactions like personal that you can

tell

us finally it's the most common reaction

um and obviously that reaction comes

from those people who have probably

tested a lot of AI tools outside of our

digital ecosystem before um but but also

we oh there have been so many questions

there there have been questions about

guidelines um there have been

journalists contacting us wondering

whether you should use this for

illustrating historical podcast pieces

uh there's been a lot of questions um

and in terms of use patterns U basically

the most the most used use case so far

seems to be chatting with your headlines

to try to improve them which I think is

quite clever from our yeah our editors

yeah

absolutely all right we're uh we're

running it's very popular we probably

could talk for a long time about this

but uh we'll we'll come back to this um

topic again I'm sure um I'm going to

share my screen

again

um so um this as I said this webinar

will be available to watch On Demand uh

on the innovate Local website uh and

we'll send out an email with that link

um there will also be a write up and

we'll have some other resources

available um if you have question any

questions about innovate local uh

suggestions for cases or maybe you have

a case that you would like to present um

please get in touch with me on nichlas

um and our next webinar is in two weeks

on March 20th and we'll feature an

exciting case from tamedia in

Switzerland the topic will be audience

development and we'll hear how Tamia

offers hyper local newsletters to

communities with as few as 5,000

inhabitants and through smart data

collection um the time to produce these

newsletters have gone from several hours

to 20 minutes uh and the newsletters now

have a reach of 20 to 40% after just one

year of of

operation and the job is to drive

Tamia's top of subscription funnel so um

now I realize audience development might

not be your main topic of interest but

it will be no doubt be very interesting

to one of or two of your colleagues so

we really recommend that you spread the

word in your companies about innovate

local um to people to come and register

and then to get on that all important um

mailing list to get the link to the

webinar every Monday and you can choose

whether you join or not so let everybody

know uh that concludes uh this innovate

local webinar thank you so much Marcus

for sharing your sites and doing a live

demo that was very brave um and thank

you for supporting our program we really

appreciate it uh and thanks everyone for

joining us today we hope to see you here

again soon for another Wednesday webinar

goodbye goodbye thank you Marcus goodbye

everyone