one all from the first leg but atalanta had just scored so now ahead on aggregate. and its also the second leg of the europa conference league olympiakos have the home advantage over aston villa. 4—2 the score from the first leg. the greeks had the advantage. it is currently 1—0. withjust the greeks had the advantage. it is currently 1—0. with just 30 the greeks had the advantage. it is currently 1—0. withjust 30 minutes gone so aston below with more work to do if they are going to get back into the sky. britain's jack draper is through to the second round of the italian open in rome, beating borna choric in straight sets. earlier in the italian capital, rafael nadal in what is expected to be his final year on the tour, made it through to the second round of the tournament after a three—set victory over zizou bergs. nadal came from a set down to win. the spaniard is a ten=time champion at the italian open and will face hubert hurkacz next. always to be emotional to play here these are the most important events

in my tennis career. the crowd have been amazing with me, supporting me since the beginning of my tennis career. the super excited to be able to play one more time here. the defending champion daniil medvedev will get his italy open campaign under way on friday and will play jack draper. medvedev was forced to pull out of the quarterfinal at the madrid open with a leg injury, but he's confident that he is ready for rome. every match was an unbelievable feeling because of the end result. honestly, i think it gave me a lot of confidence for this year. but where this is the first of second year, together with last year, where i am ok to play on clay. i don't really hate it. i know there are some parts of my game which are may be a not adopted 100%, but i know i can win the big players, i know i

can win the big players, i know i can win the big players, i know i can win big tournaments on clay. i am feeling good about it and it helps a lot for the mental part. naomi osaka put in one of her best performances of the season after beating marta kostyuk in straight sets in the second round in rome. the win against kostyuk is the first time osaka has won two matches in a row on clay since rome since 2019, when she reached the quarterfinals. she won the rain—interrupted match in 71 minutes. pelayo sanchez beat julian alaphilippe and luke plapp to the line on stage six of the giro d'italia. as the race wound its way through the white roads of tuscany, after trying for many kilometres, the break finally went away and managed to get a lead over the chasing peloton. but it came down to three men for the sprint into rapolano terme. and the 24—year—old sanchez had the legs to keep the others at bay and cross the line in the lead. tadej pogacar finished safely in the peeton to retain the race leader's pinkjersey. and that's all the sport for now.

you are watching the context. it is time for our regular weekly segment, ai decoded. welcome to ai decoded. every week in this programme, we take you deep into the world of artificial intellligence, and what we have tried to do is focus these programmes on one particular theme. last week, we looked at the advance of al on the battlefield. tonight, we are going to consider something a number of you have raised, and that is al and energy, or more specifically the energy that al consumes. there are climate experts who are warning that the advance of artificial intelligence could lead to an 80% increase in our global carbon emissions. let's start with sustainability, because that cloud that al models live on is actually made out of metal, plastic and powered by vast amounts of energy.

and each time you query an ai model, it comes with the cost to the planet. that is sasha luccioni, lead climate researcher at the ai company hugging face. she's going to be with us from montreal injust a moment. how much energy are we using? according to this report from vox, ai is already consuming as much energy as a small country and we are only at the beginning. the next web says this is where our two existential crises collide with one another, the climate crisis and the exponential growth of ai. can one help solve the other, or will it exacerbate the problem? here in the studio, our regular ai contributor priya lakhani, ceo of the ai—powered education company century tech. welcome. good to see you. when it comes to di . ital good to see you. when it comes to digital everyone — good to see you. when it comes to digital everyone knows _ good to see you. when it comes to digital everyone knows there - good to see you. when it comes to digital everyone knows there is - good to see you. when it comes to digital everyone knows there is a i digital everyone knows there is a cost, the wiring, chips, precious metals, the water that cools the data processing centres. what we

don't often talk about is the energy going into ai and the training of large language marbles like chat gpt. �* , , large language marbles like chat gpt. , a, ., gpt. because they are far more hunu gpt. because they are far more hungry that _ gpt. because they are far more hungry that we _ gpt. because they are far more hungry that we have _ gpt. because they are far more hungry that we have seen - gpt. because they are far more i hungry that we have seen before. gpt. because they are far more - hungry that we have seen before. we have talked about air models and played with them. they are trained and deployed india centres. the data centres consume vast amounts of electricity. so if you are powering the data centre with nonrenewable sources, then, essentially you have potentially huge carbon emissions and those particular models we are talking about the generative ai ai models, but also the models, images videos consume a large amount. so in 2019 we said stream in our video it was 36 grams of catjust to put that into context for everyone because thatis into context for everyone because that is why we are here, that is driving a car typical petrol car about 160 metres. meta had just

relieved the third model on one of my favourite platforms and they said that admitted about 2290 metric tonnes of co2. so to put that into context, that is about 500 average cars and what they admit in an entire year. so you start to get what we had with digital which is what we had with digital which is what you asked all the way to these really huge ai models. i want to show you what i looked a little bit earlier before i came in. ijust looked at half of an hour of the images produced on major knee. this is what you're seeing now. so you have about ten unique images produced by this particular platform about half an hour about 430 uk time today. just as images. just on one channel of where you can find it. there are hundreds and hundreds, but just those ten along with take about two and half times the battery charge of the phone. so this is a

significant problem and if anyone is interested there are great interviews with mark zuckerberg and others who said there is going to be an energy bottleneck. obviously for activists this is a significant problem. some really looking forward to sasha. let's bring in sasha luccioni. as the lead climate researcher at hugging face, she's spent nearly a decade looking at data storage and machine learning and how this all contributes to our energy consumption. welcome to the programme. thank you for having me. before we get started, maybe i could frame our conversation with an image you sent us that visuallyjust underlines how expensive our digital usage has become. so here is google's annual energy use, 18.3 trillion watts, 10—15% of that currently going towards ai. and here is what the republic of ireland uses in any one year, 29.3 trillion watts. so one company, sasha, is now consuming two thirds of what a small company

uses each year. yet, i don't hear governments talking about that problem. i think that is because _ talking about that problem. i think that is because ai _ talking about that problem. i think that is because ai is _ talking about that problem. i think that is because ai is a _ talking about that problem. i think that is because ai is a horizontal i that is because ai is a horizontal it is not— that is because ai is a horizontal it is not agricultural transportation and it affects all industries. anything that uses ai from _ industries. anything that uses ai from navigation to web search so i think_ from navigation to web search so i think governments don't know what bucket_ think governments don't know what bucket to _ think governments don't know what bucket to put it in and when you don't _ bucket to put it in and when you don't know. _ bucket to put it in and when you don't know, you lent it slip through the cracks — don't know, you lent it slip through the cracks. can don't know, you lent it slip through the cracks. . , ., don't know, you lent it slip through the cracks. ., , ., ,. , don't know, you lent it slip through the cracke— the cracks. can you describe this in the cracks. can you describe this in the terms of _ the cracks. can you describe this in the terms of large _ the cracks. can you describe this in the terms of large language - the cracks. can you describe this in i the terms of large language models, what their usage is like and actually tell us what that means in terms of energy consumption and the impact on the environment. language models have — impact on the environment. language models have become _ impact on the environment. language models have become one _ impact on the environment. language models have become one of - impact on the environment. language models have become one of the - impact on the environment. language models have become one of the mostj models have become one of the most popular— models have become one of the most popular usages of ai and they are being _ popular usages of ai and they are being deployed in everything nowadays. you can talk to your stove or your _ nowadays. you can talk to your stove or your fridge — nowadays. you can talk to your stove or your fridge and in a recent study we found _ or your fridge and in a recent study we found that training a large language model is very energy intensive. that is the numbers you

give _ intensive. that is the numbers you give. actually each query uses energy — give. actually each query uses energy. depending on the size of the model. _ energy. depending on the size of the model, zoo-500,000,000 energy. depending on the size of the model, zoo—500,000,000 queries will equal the _ model, zoo—500,000,000 queries will equal the amount of energy for training — equal the amount of energy for training. so it may seem like a lot, but for— training. so it may seem like a lot, but for chatgpt, it averages around 10 million— but for chatgpt, it averages around 10 million users a day. so within a couple _ 10 million users a day. so within a couple of— 10 million users a day. so within a couple of weeks you have this vast amount— couple of weeks you have this vast amount of— couple of weeks you have this vast amount of energy that is equivalent to all— amount of energy that is equivalent to all of— amount of energy that is equivalent to all of these cars over a year. butiust— to all of these cars over a year. butjust with people using the to all of these cars over a year. but just with people using the tool. did you _ but just with people using the tool. did you see the story this week that microsoft are going to plough in about $100 billion into this supercomputer called stargate and it will be powered by not one, they say, but several nuclear power stations. that got me thinking because i have heard the head at the talk about this and he said, that is how we will have to work, we will have to create our own energy systems. is that where new energy comes from, the biggest companies in the world driving investment? it is

a problem- — the world driving investment? it is a problem. because to what extent do you want— a problem. because to what extent do you want big tech companies building their own _ you want big tech companies building their own nuclear reactors? may be that energy — their own nuclear reactors? may be that energy could be better used for other— that energy could be better used for other things. that energy could be better used for otherthings. because that energy could be better used for other things. because we should be decarbonise in our energy globally. currently. — decarbonise in our energy globally. currently, if we are going to funnel all of— currently, if we are going to funnel all of that— currently, if we are going to funnel all of that investment into the energy— all of that investment into the energy use for al may be other sectors — energy use for al may be other sectors will get overlooked and mayhe — sectors will get overlooked and maybe we should focus on those if you want _ maybe we should focus on those if you want to — maybe we should focus on those if you want to decarbonise. what i'm excited about _ you want to decarbonise. what i'm excited about is _ you want to decarbonise. what i'm excited about is when _ you want to decarbonise. what i'm excited about is when we - you want to decarbonise. what i'm excited about is when we talk - you want to decarbonise. what i'ml excited about is when we talk about al models and i was on hugging face ai models and i was on hugging face today, for those who do not know what it is it is a marker one model repository i like models have a description now on potentially how much energy they have used. you produce something novel at hugging face so when you are building these ai models we are focused on the latency and speed and high performance the model is but you are potentially creating a little trip adviser on how efficient is. tell us about that. adviser on how efficient is. tell us about that-— adviser on how efficient is. tell us about that. , , ., , ., about that. currently when people go on hu: tlnt about that. currently when people go on hugging face _ about that. currently when people go on hugging face they _ about that. currently when people go on hugging face they shop _ about that. currently when people go on hugging face they shop around i about that. currently when people go | on hugging face they shop around for

models _ on hugging face they shop around for models that work for the task they want to— models that work for the task they want to do — models that work for the task they want to do. it could be language, audio. _ want to do. it could be language, audio. it— want to do. it could be language, audio, it could be image generation or even_ audio, it could be image generation or even looking at video. typically they look— or even looking at video. typically they look at things like performance or latency _ they look at things like performance or latency but i'm calculating the energy— or latency but i'm calculating the energy usage across all different tasks _ energy usage across all different tasks in — energy usage across all different tasks in models on the hub, hugging face and _ tasks in models on the hub, hugging face and to — tasks in models on the hub, hugging face and to provide that information to people _ face and to provide that information to people so they can factor it into maybe _ to people so they can factor it into maybe it _ to people so they can factor it into maybe it is — to people so they can factor it into maybe it is not only faster but more efficient _ maybe it is not only faster but more efficient but must performative but vastly— efficient but must performative but vastly more efficient. signs developing energy star ratings for al models. ,, . ., ., ,, ., ai models. since we are talk about how to mitigate _ ai models. since we are talk about how to mitigate the _ ai models. since we are talk about how to mitigate the problem. - ai models. since we are talk about how to mitigate the problem. let l let me introduce you to chris starkey. he is the ceo of the london—based start—up nexgen cloud. they have been in business since 2020, sourcing data centres that are entirely powered by renewable energy. welcome to the programme. tell us what you do and how your clients would typically work. we what you do and how your clients would typically work.— what you do and how your clients would typically work. we are on the other side of— would typically work. we are on the other side of the _ would typically work. we are on the other side of the fence. _ would typically work. we are on the other side of the fence. our - other side of the fence. our business _ other side of the fence. our business is _ other side of the fence. our business is all— other side of the fence. our business is all about - other side of the fence. ourl business is all about building other side of the fence. our- business is all about building large

scale _ business is all about building large scale gpu — business is all about building large scale gpu clusters. _ business is all about building large scale gpu clusters. basically- business is all about building large scale gpu clusters. basically we i scale gpu clusters. basically we have _ scale gpu clusters. basically we have a _ scale gpu clusters. basically we have a core _ scale gpu clusters. basically we have a core focus _ scale gpu clusters. basically we have a core focus on _ scale gpu clusters. basically we have a core focus on building i have a core focus on building identity— have a core focus on building identity accelerated - have a core focus on building i identity accelerated compute. this is typically — identity accelerated compute. this is typically what _ identity accelerated compute. this is typically what companies, - identity accelerated compute. this is typically what companies, may. identity accelerated compute. this i is typically what companies, may be that the _ is typically what companies, may be that the use — is typically what companies, may be that the use hugging _ is typically what companies, may be that the use hugging face _ is typically what companies, may be that the use hugging face or- that the use hugging face or companies _ that the use hugging face or companies using _ that the use hugging face or companies using or- that the use hugging face or companies using or buildingl that the use hugging face or- companies using or building their own foundational— companies using or building their own foundational models, - companies using or building their own foundational models, this i companies using or building their own foundational models, this is| companies using or building their. own foundational models, this is the type of— own foundational models, this is the type of infrastructure _ own foundational models, this is the type of infrastructure that _ own foundational models, this is the type of infrastructure that they i type of infrastructure that they would — type of infrastructure that they would be — type of infrastructure that they would be consuming. - type of infrastructure that they would be consuming. our- type of infrastructure that they i would be consuming. our mission is to deliver— would be consuming. our mission is to deliver at — would be consuming. our mission is to deliver at scale _ would be consuming. our mission is to deliver at scale 100% _ would be consuming. our mission is to deliver at scale 100% new- to deliver at scale 100% new renewable _ to deliver at scale 100% new renewable he _ to deliver at scale 100% new renewable he powered. i to deliver at scale 100% new renewable he powered. hourj to deliver at scale 10096 new renewable he powered. how do we achieve this _ renewable he powered. how do we achieve this optimisation _ renewable he powered. how do we achieve this optimisation of - renewable he powered. how do we l achieve this optimisation of models? are also to things we could be looking out. intra— structure, efficiency, and i know you look at in terms of cooling technology, i would love to hear more on that. also the optimised hardware, for example what nvidia brought out with blackwell saying that it reduces cost for companies. so can you describe these methodologies, how reliant on these and when we talk about cooling methodologies, explain

that. we know the data centres and these centres need cooling but walk us through it. hope these centres need cooling but walk us through it— these centres need cooling but walk us through it. how long have we got. i'm not us through it. how long have we got. i'm not too — us through it. how long have we got. i'm not too sure. _ us through it. how long have we got. i'm not too sure. not— us through it. how long have we got. i'm not too sure. not long. _ us through it. how long have we got. i'm not too sure. not long. at i us through it. how long have we got. i'm not too sure. not long. at a i i'm not too sure. not long. at a hirh i'm not too sure. not long. at a high level. — i'm not too sure. not long. at a high level, only _ i'm not too sure. not long. at a high level, only a _ i'm not too sure. not long. at a high level, only a few— i'm not too sure. not long. at a high level, only a few years i i'm not too sure. not long. at a high level, only a few years ago i'm not too sure. not long. at a i high level, only a few years ago we were _ high level, only a few years ago we were building — high level, only a few years ago we were building high—performance i were building high—performance environments _ were building high—performance environments that _ were building high—performance environments that may- were building high—performance environments that may be i were building high—performancel environments that may be 10—20 kilowatts — environments that may be 10—20 kilowatts was _ environments that may be 10—20 kilowatts was deemed _ environments that may be 10—20 kilowatts was deemed as - environments that may be 10—20 kilowatts was deemed as high. . environments that may be 10—20 i kilowatts was deemed as high. now, quite commonly— kilowatts was deemed as high. now, quite commonly we _ kilowatts was deemed as high. now, quite commonly we are _ kilowatts was deemed as high. now, quite commonly we are building i kilowatts was deemed as high. now, quite commonly we are building outi quite commonly we are building out environments— quite commonly we are building out environments that _ quite commonly we are building out environments that are _ quite commonly we are building out environments that are 50—60 - environments that are 50—60 kilowatts _ environments that are 50—60 kilowatts and _ environments that are 50—60 kilowatts and the _ environments that are 50—60 kilowatts and the new- environments that are 50—60 i kilowatts and the new iteration, environments that are 50—60 - kilowatts and the new iteration, the next generation _ kilowatts and the new iteration, the next generation of— kilowatts and the new iteration, the next generation of chips _ kilowatts and the new iteration, the next generation of chips and - kilowatts and the new iteration, the next generation of chips and some i kilowatts and the new iteration, the i next generation of chips and some of the infrastructure _ next generation of chips and some of the infrastructure that _ next generation of chips and some of the infrastructure that we _ next generation of chips and some of the infrastructure that we are - the infrastructure that we are bringing _ the infrastructure that we are bringing into— the infrastructure that we are bringing into play— the infrastructure that we are bringing into play for- the infrastructure that we are bringing into play for next i the infrastructure that we are. bringing into play for next year the infrastructure that we are i bringing into play for next year for 2025, _ bringing into play for next year for 2025, would — bringing into play for next year for 2025, would have _ bringing into play for next year for 2025, would have been _ bringing into play for next year for 2025, would have been north- bringing into play for next year for 2025, would have been north of. bringing into play for next year for i 2025, would have been north of 120 kilowatts _ 2025, would have been north of 120 kilowatts per — 2025, would have been north of 120 kilowatts per rack. _ 2025, would have been north of 120 kilowatts per rack. so _ 2025, would have been north of 120 kilowatts per rack. so we _ 2025, would have been north of 120| kilowatts per rack. so we are seeing a clear— kilowatts per rack. so we are seeing a clear increase _ kilowatts per rack. so we are seeing a clear increase in _ kilowatts per rack. so we are seeing a clear increase in its _ kilowatts per rack. so we are seeing a clear increase in its financial- a clear increase in its financial growth — a clear increase in its financial growth in _ a clear increase in its financial growth in density. _ a clear increase in its financial growth in density. that's i a clear increase in its financial. growth in density. that's great. a clear increase in its financial- growth in density. that's great. we can fit— growth in density. that's great. we can fit more — growth in density. that's great. we can fit more power— growth in density. that's great. we can fit more power into _

growth in density. that's great. we can fit more power into data - growth in density. that's great. wel can fit more power into data centres but they— can fit more power into data centres but they eventually _ can fit more power into data centres but they eventually will _ can fit more power into data centres but they eventually will be - can fit more power into data centres but they eventually will be having i can fit more power into data centres but they eventually will be having a i but they eventually will be having a hu-e but they eventually will be having a huge amount — but they eventually will be having a huge amount of— but they eventually will be having a huge amount of resource, - but they eventually will be having a huge amount of resource, energy. i but they eventually will be having a . huge amount of resource, energy. so it isjust _ huge amount of resource, energy. so it is just unsustainable _ huge amount of resource, energy. so it is just unsustainable to _ huge amount of resource, energy. so it is just unsustainable to have - it isjust unsustainable to have traditional— it isjust unsustainable to have traditional techniques- it isjust unsustainable to have traditional techniques of- it is just unsustainable to have i traditional techniques of cooling like air— traditional techniques of cooling like air cooling _ traditional techniques of cooling like air cooling which _ traditional techniques of cooling like air cooling which is - traditional techniques of cooling like air cooling which is quite i like air cooling which is quite common— like air cooling which is quite common now _ like air cooling which is quite common now. so— like air cooling which is quite common now. so we - like air cooling which is quite common now. so we have i like air cooling which is quite i common now. so we have a keen like air cooling which is quite _ common now. so we have a keen focus on trialing _ common now. so we have a keen focus on trialing and — common now. so we have a keen focus on trialing and testing _ common now. so we have a keen focus on trialing and testing new _ common now. so we have a keen focus on trialing and testing new ways - common now. so we have a keen focus on trialing and testing new ways to - on trialing and testing new ways to cool the _ on trialing and testing new ways to cool the chips _ on trialing and testing new ways to cool the chips. one _ on trialing and testing new ways to cool the chips. one of— on trialing and testing new ways to cool the chips. one of the - on trialing and testing new ways to cool the chips. one of the new- on trialing and testing new ways toi cool the chips. one of the new ways is something — cool the chips. one of the new ways is something called _ cool the chips. one of the new ways is something called liquid _ cool the chips. one of the new ways is something called liquid core - is something called liquid core directed — is something called liquid core directed chip. _ is something called liquid core directed chip. this _ is something called liquid core directed chip. this brings- is something called liquid core directed chip. this brings a - is something called liquid core l directed chip. this brings a huge amount— directed chip. this brings a huge amount of— directed chip. this brings a huge amount of efficiency— directed chip. this brings a huge amount of efficiency but - directed chip. this brings a huge amount of efficiency but at - directed chip. this brings a huge amount of efficiency but at the l directed chip. this brings a huge - amount of efficiency but at the same time, _ amount of efficiency but at the same time, as_ amount of efficiency but at the same time, as chips— amount of efficiency but at the same time, as chips get— amount of efficiency but at the same time, as chips get more _ amount of efficiency but at the same time, as chips get more powerful, . amount of efficiency but at the samel time, as chips get more powerful, we are obviously— time, as chips get more powerful, we are obviously drying _ time, as chips get more powerful, we are obviously drying up _ time, as chips get more powerful, we are obviously drying up a _ time, as chips get more powerful, we are obviously drying up a huge - are obviously drying up a huge amount— are obviously drying up a huge amount more _ are obviously drying up a huge amount more power— are obviously drying up a huge amount more power per- are obviously drying up a huge l amount more power per square are obviously drying up a huge - amount more power per square feet, and each — amount more power per square feet, and each data _ amount more power per square feet, and each data centre. _ amount more power per square feet, and each data centre. [— amount more power per square feet, and each data centre.— and each data centre. i was going to say because — and each data centre. i was going to say because we _ and each data centre. i was going to say because we are _ and each data centre. i was going to say because we are pressed - and each data centre. i was going to say because we are pressed for - and each data centre. i was going to| say because we are pressed for time. just a final answer from both of you and maybe you can chip in on this. pardon the pun. what about sovereignty? everyone wants control of their own computers and obviously some of that is going to come down to where the cloud is, what energy they have, what energy they can generate. if we are going to make this available to everybody, how concerned are you both by that? i

concerned are you both by that? i feel ai is really slipping through the cracks when it comes to accounting for energy and carbon because — accounting for energy and carbon because it — accounting for energy and carbon because it is often companies in one country— because it is often companies in one country using cloud commuting in another— country using cloud commuting in another country and often for example _ another country and often for example everyone i talk to cloud providers — example everyone i talk to cloud providers they say, we don't know what _ providers they say, we don't know what is _ providers they say, we don't know what is running on our centres it could _ what is running on our centres it could be — what is running on our centres it could be streaming or ai so it is hard _ could be streaming or ai so it is hard for— could be streaming or ai so it is hard for them to account for this energy— hard for them to account for this energy usage. so every time i say, -ive energy usage. so every time i say, give me — energy usage. so every time i say, give me a _ energy usage. so every time i say, give me a number, they say we don't have them _ give me a number, they say we don't have them. so it is currently not being _ have them. so it is currently not being accounted for, let's say. if being accounted for, let's say. they're being accounted for, let's say. if they're trying to do it sustainably ithink— they're trying to do it sustainably ithihk a — they're trying to do it sustainably ithihk a tot — they're trying to do it sustainably i think a lot of— they're trying to do it sustainably i think a lot of countries - they're trying to do it sustainably i think a lot of countries will - i think a lot of countries will struggle _ i think a lot of countries will struggle they— i think a lot of countries will struggle. they absolutely. i think a lot of countries will . struggle. they absolutely will. there's — struggle. they absolutely will. there's enough _ struggle. they absolutely will. there's enough infrastructure i there's enough infrastructure tocatty — there's enough infrastructure locally to _ there's enough infrastructure locally to provide _ there's enough infrastructure locally to provide sustainable| locally to provide sustainable infrastructure. _ locally to provide sustainable infrastructure. not _ locally to provide sustainable infrastructure. not at - locally to provide sustainable infrastructure. not at the - locally to provide sustainable i infrastructure. not at the scale intimate — infrastructure. not at the scale intimate we _ infrastructure. not at the scale intimate we are _ infrastructure. not at the scale intimate we are seeing - infrastructure. not at the scale i intimate we are seeing currently infrastructure. not at the scale - intimate we are seeing currently for stuff every — intimate we are seeing currently for stuff every country _ intimate we are seeing currently for stuff every country is _ intimate we are seeing currently for stuff every country is going - intimate we are seeing currently for stuff every country is going to - intimate we are seeing currently for stuff every country is going to want| stuff every country is going to want a sovereign — stuff every country is going to want a sovereign cloud, _ stuff every country is going to want a sovereign cloud, they— stuff every country is going to want a sovereign cloud, they all- a sovereign cloud, they all absolutely _ a sovereign cloud, they all absolutely going _ a sovereign cloud, they all absolutely going forward l a sovereign cloud, they all. absolutely going forward right a sovereign cloud, they all- absolutely going forward right now and everyone _ absolutely going forward right now and everyone went _ absolutely going forward right now and everyone went their— absolutely going forward right now and everyone went their own - and everyone went their own sovereign _ and everyone went their own sovereign gpt, _ and everyone went their own sovereign gpt, they- and everyone went their own sovereign gpt, they won't i and everyone went their own| sovereign gpt, they won't be and everyone went their own - sovereign gpt, they won't be able to do it currently— sovereign gpt, they won't be able to do it currently certainly— sovereign gpt, they won't be able to do it currently certainly not - sovereign gpt, they won't be able to do it currently certainly not in - sovereign gpt, they won't be able to do it currently certainly not in the - do it currently certainly not in the

uk. , ., ., , ., do it currently certainly not in the uk. .,,, do it currently certainly not in the uk. everyone was a super computer she wasjust — uk. everyone was a super computer she was just saying. _ uk. everyone was a super computer she wasjust saying. i _ uk. everyone was a super computer she wasjust saying. i was _ uk. everyone was a super computer she wasjust saying. i wasjust - she was just saying. i was just thinking back to 620 to the climate conferences and when people talk about carbon footprints and what belongs to the carbon footprint. we talk about the missions but we never talk about the missions but we never talk about the missions but we never talk about cloud power or computer junta reading. it is talk about cloud power or computer junta reading-— junta reading. it is the length of time for sunply _ junta reading. it is the length of time for supply to _ junta reading. it is the length of time for supply to create - junta reading. it is the length of time for supply to create these | time for supply to create these renewable energy data centres to 100% of demand. think about how quick it was that chat6pt exploded. amazing. amazing stuff and thank you for being with us here on ai6 coated. after the break preamble guide is through some of the big stories for the week. has anyone seen the ad for the apple ipad and we will see an ad made entirely by ai we will see an ad made entirely by aland how long before we can speak to animals with mac we will be right

back.

welcome back. the new ad for the ai—powered ipad pro has sparked quite the backlash among hollywood creatives. we will find out why. maybe it is because they can now create an entire movie without auditioning someone who looks like the russian president. this is the ai—generated putin, a biopic the guardian says will be out for release in september. if you can reproduce vladimir putin, then it stands to reason you can recreate anyone, including the dead. the times reports there's been a surge in deadbots, or griefbots. families using al to bring their loved one back to life. but does it require tighter regulation? and how long before we can understand what the animals are saying? scientists that have spoken to sky news say they think ai will one day help them communicate with a sperm whale. dr dolittle, i presume! so, priya, there is already huge

sensitivity within the creative arts industries that al is replacing the talent, is destroying the human experience of playing an instrument or writing a song. so with that in mind, let me play you the new ad from apple that's been released to promote the new ai ipad pro. music: all i ever need is you by sonny & cher # sometimes when i'm down and all alone #0h... # all i ever need is you # winters come and they go # and we watch the melting snow # so as summer follows spring...# it is called crush. "just imagine all the things it'll be used to create," said apple ceo tim cook. let me read you the response from

director and writer asif kapadia. "don't know why anyone thought this ad was a good idea. it is the most honest metaphor for what tech companies do to the arts, to musicians, creators, writers, film—makers — squeeze them, use them, not pay well, take everything, then say it's all created by them." we have talked about this on the programme before. ads, we have talked about this on the programme before. fix. bit we have talked about this on the programme before. a bit insensitive? yes there is — programme before. a bit insensitive? yes there is anger _ programme before. a bit insensitive? yes there is anger because _ programme before. a bit insensitive? yes there is anger because of- programme before. a bit insensitive? yes there is anger because of the - yes there is anger because of the threat ai poses but also the idea that tech really misunderstands the art in the fact that what is technology's role in diminishing artistic and cultural mediums. what is really interesting and i think there is a funny video that can we play it will take 30 seconds on actually what one director created in response to apple's adverts. can we get that up? it was a film—maker who reversed it and essentially describing, a good quality video essentially all of the arts,

instruments crushing the ipad in the reverse of that. why do i not think that this is an actual threat? because instruments are notjust tools. your guitar, piano, it is a tactile immersive experiments most i have a neck on ds a lark it enables me to have that artistic creativity when i am taking a picture. in ipad, as great as it is, i love my ipad, but it is not able to do. when it comes to painting and drawing can be of artists that use difficult physical mediums like oils and acrylics, it is not all about that physical. acrylics, it is not all about that -h sical. ., , acrylics, it is not all about that -h sical. .,, . ., acrylics, it is not all about that -h sical. ., , . ., . physical. people are wondering what the movies of— physical. people are wondering what the movies of the _ physical. people are wondering what the movies of the future _ physical. people are wondering what the movies of the future might - physical. people are wondering what the movies of the future might look| the movies of the future might look like and what it might do to the cinematic arts in particular. maybe ship a quick look at this. this is the new ai biopic entirely created by artificial intelligence. the president has found time for you after all. this will culminate in a nuclear strike.

only i can save you and your family from prison. vladimir putin has won i with 53.44% of the vote. it's an embarrassing result. what would have been the satisfactory result for you, mr president? 100. what is interesting about that and what panics people as you don't need auditions for people that look like putin they can create him. they have a bill clinton there that looks like bill clinton. there may be an issue over that. . , ,., , , clinton. there may be an issue over that. . , , , ., clinton. there may be an issue over that. , , ,., that. absolutely in terms of your ima . e that. absolutely in terms of your image rights _ that. absolutely in terms of your image rights and _ that. absolutely in terms of your image rights and we _ that. absolutely in terms of your image rights and we have - that. absolutely in terms of your image rights and we have talked | image rights and we have talked about the legal issues before. but you can do this at low cost with the sort of technology now. you don't have to deal with the egos, necessarily that you may otherwise have to. they are huge ethical concerns, notjust because it is easy talking about putin. a lot of the world may say that is fine, but actually were starting to blur the lines between reality and fiction. you can misrepresent, in many ways.

you can misrepresent, in many ways. you question in why theyjust did not have an actor in the biopic. this is clearlyjust about using that technology. they're so much more to come in the future when we talk about the peaks and whether they are a good or bad use. clearly, if this was that you or of rishi sunak or keir starmer, then it would be up in arms about it. == be up in arms about it. -- deepfakes. _ be up in arms about it. -- deepfakes. it _ be up in arms about it. —— deepfakes. it takes a much greater effect when it looks specifically like them. the best story of the week is this story about sperm whales. we have known for some time that their calls, their clicks, are highly sophisticated to coordinte and communicate with one another. but what if there is actually a language? the scientiests studying these whales in the east caribbean found they use morse code with a rhythm and a tempo that suggests it is conveying meaning. and that's where ai comes in. potentially... how? when you are lookin: at potentially... how? when you are looking at language, _ potentially... how? when you are looking at language, will - potentially... how? when you are looking at language, will be - potentially... how? when you are i looking at language, will be mapped language, we map the words. imagine

creating a map of all of the words of where they are. then you catholic the statistical difference between words. this is how you end up with these large language models —— you calculate. here, what you are looking at is when you have these clicks, they are looking simplistically for patterns between them. and they are recording what them. and they are recording what the wales are doing and finding how those patterns might relate to diving really deep or coming up or feeding and then they are attaching essentially those patterns and creating this map and language and then, there will be an exercise where they will try to essentially overlay that with the english language. the test is when they hear these clicks, really loud, by the way. if you watched videos on these wales they are super loud. but when they hear those clicks, can the ai model that they create can then predict with the whale is going to do. if predict with the whale is going to do. ., a

predict with the whale is going to do. ., , ., predict with the whale is going to do. ., . do. if it works then you connect them a bit _ do. if it works then you connect them a bit further _ do. if it works then you connect them a bit further and - do. if it works then you connect them a bit further and start - do. if it works then you connect them a bit further and start to i them a bit further and start to create the language and talk back. people... this raises huge ethical issues on how we are interacting with animals. and whether we should stop should you? i and whether we should stop should ou? ., �* . , ., and whether we should stop should ou? .,�* . , ., and whether we should stop should ou? ., �* . , ., . ., you? i don't want my dog asking for a biscuit every _ you? i don't want my dog asking for a biscuit every five _ you? i don't want my dog asking for a biscuit every five minutes. - a biscuit every five minutes. absolutely not. talk to the i don't want my dog telling me he prefers my husband.— i don't want my dog telling me he prefers my husband. that's the worst one. the prefers my husband. that's the worst one- the thing _ prefers my husband. that's the worst one. the thing is, _ prefers my husband. that's the worst one. the thing is, obviously - prefers my husband. that's the worst one. the thing is, obviously there's l one. the thing is, obviously there's a lot we can learn as well. but there's going to be huge questions about it. there've been lots of people working on that's and elephants and there has been a fantastic podcast done on it and i suggest you listen to it. congratulation to sky news who picked that up in the scientists. that is it from us. same time next week. all the programmes we do are

on the bbc�*s youtube channel. all the big interviews, lots of good people coming on the programme now so if you want to look at previous episodes, go and have a look at that. hello there. temperatures have been creeping up day by day. high pressure has brought plenty of sunshine around, but a weather front launching across northern scotland has brought thicker cloud here, some spots of rain and thursday was another grey day here with outbreaks of rain. now, as we move through tonight, it does look like that weather front will eventually move northwards, become confined to the northern isles for a while through the night before it clears away. it becomes drier here, but a dry night to come for most areas. bit of mist and fog here and there, temperatures ranging from 9—12 celsius. friday, then, our area of high pressure continues to bring a lot of fine and settled weather. we lose that weather front from northern scotland, so, apart from a bit of early cloud, some mist and fog, most places should have a fine dry day. widespread sunshine, particularly for england and wales, into northern ireland, central and southern

scotland, but some sunshine into the highlands as well. that'll lift temperatures up to 22 celsius perhaps in aberdeen. 23—24, perhaps, in the warmest spots in england and wales. through friday night, it looks like we'll see mist and fog returning in places. a bit of low cloud, sea fog pushing into eastern england, east anglia and the south—east. that could creep a little bit further westwards into the midlands first thing on saturday. but again, those temperatures, 9—12 celsius. now for the weekend. it does stay warm, there will be some good spells of sunshine around, but the shower risk will start to increase, particularly as we head into sunday. that's because we've got a weather front and low pressure developing, pushing towards our shores, particularly by sunday. and into next week, it'll be a lot more unsettled than we've had this week. saturday, though, another warm day to come, a bit of early mist and fog and some low cloud across eastern england. otherwise, plenty of sunshine, though we could see a few heavy showers develop across the high ground of northern england, central and southern scotland into the afternoon. that could be thundery as well.

but a warm day to come, quite humid too, the low—to—mid—205 quite widely. sunday probably the warmest day of the week across southern areas. it could be up to 26 celsius, plenty of sunshine. but then the showers and thunderstorms could become a bit more widespread through the afternoon. and the temperatures may be coming down a little bit across western areas as the clouds build up in the sky. then as we head into next week low pressure takes over, it will become a lot more unsettled, showers or longer spells of rain at times. but there will still be a little bit of sunshine too, take care.

hello, i'm christian fraser. you're watching the context on bbc news. the us fears that if israel goes ahead with a full invasion of rafah, there will be carnage. already there's fear and panic with tens of thousands fleeing, and food and fuel quickly running out. tragicallyjoe biden has been the greatest friend to hamas and hezbollah that there is on planet earth. now those sound like extraordinary statements. what are the facts? i think within the administration, they believe that israel has - everything it needs in terms of american armaments. . if it wanted to go into rafah, it could still do that. - this is about trying to pressure mr netanyahu into changing i or modifying his decision.

joining me tonight are leslie vinjamuri, the director of the us & americas at the chatham house, and republican strategist mike shields in washington. we will get to mike and leslie very shortly. first, the latest headlines — president biden has warned benjamin netenyahu that the us will stop more weapons bound for israel if it launches a major ground operation in the gaza city of rafah. despite the concerns expressed by the white house, israel appears ready to mount a large—scale assault on the city, which they say is hamas's last major stronghold in the territory. thousands of people have joined pro—palestinian protests a man has been arrested on suspicion of murder after a woman in her 605 was fatally stabbed in north london. paramedics and an air ambulance

attended the incident and treated the woman for stab injuries, but she died at the scene.