Magna Concursos

Foram encontradas 46.411 questões.

Space power: The dream of beaming solar energy from orbit

Harvesting solar energy in orbit and beaming it down to Earth is a decades-old idea. Now,

a growing number of companies say they could finally make it a reality.

At an American football stadium in Florida last March, an unusual test took place. It wasn’t

footballs being thrown, but beams of light fired across the length of the pitch. The concentrated

streaks of light lasted for a few minutes, sent from an emitter on one side of the Jacksonville

Jaguars’ stadium and collected on a screen on the other.

The light had been collected from the Sun and then beamed out by large lenses on the field,

each about 1.2 metres (4ft) tall, acting like magnifying glasses. “We had to get up a ladder to

pull the cover off”, said Andrew Rush, chief executive of the Florida-based company Star Catcher,

which carried out the test. “We beamed 100 watts about 105 metres”.

The goal was simple: to see if sunlight could be beamed across space to power satellites or

even cities on Earth.

Space-based solar power aims to capture sunlight in orbit and _______________ it to the

ground as clean, renewable energy. Solar panels on Earth are limited by the atmosphere, the

weather and the day-night cycle. But in space, sunlight can be collected almost constantly, at

much higher _______________.

He recalled telling his father about the idea, who thought it sounded rather unrealistic. Yet,

as engineer David Homfray explained, “Space-based solar power makes the energy transition

work”. Some estimates suggest it could one day provide up to 80% of Europe’s renewable energy

needs.

Still, making it work is no small task. It would require enormous satellite structures,

complex operations and countless rocket launches. And there are cheaper renewable solutions

that could come online far faster — a crucial point if the world is to replace fossil fuels in time to

limit global warming.

The concept itself is not new. Science-fiction writer Isaac Asimov imagined it in 1941, and

NASA studied it in the 1970s and 1990s. Back then, it was dismissed as too costly and technically

unrealistic. But modern advances in rocket design, robotics and energy transmission have made

it more feasible.

Today, several countries — including the US, the UK, Japan and China — are investing in

prototypes. One approach involves using infrared lasers to beam power to the ground, while

another envisions vast solar stations in orbit that could supply electricity to entire countries.

However, not everyone agrees it’s viable. Critics warn about the risks of increasing space

_______________ and the difficulty of managing thousands of satellites safely. Others question

whether such huge orbital structures would even be legal under international space treaties.

NASA’s recent reports say the technology remains more expensive than terrestrial

renewables, yet progress is undeniable. “Once you make that initial investment”, said Michelle

Hanlon, a space-law expert, “that power is literally free”.

Whether space-based solar power will ever become practical remains uncertain. But for

now, it represents a bold vision of how humanity might one day capture sunlight beyond the

atmosphere — and beam power directly from the stars.

(Available at: www.bbc.com/future/article/20251029-the-beam-dream-should-we-build-solar-farms-in-space– text specially adapted for this test).

Mark the alternative that fills in, correctly and respectively, the blanks in the text in lines 13, 16 and 33 according to standard spelling rules.

 

Provas

Questão presente nas seguintes provas

Space power: The dream of beaming solar energy from orbit

Harvesting solar energy in orbit and beaming it down to Earth is a decades-old idea. Now,

a growing number of companies say they could finally make it a reality.

At an American football stadium in Florida last March, an unusual test took place. It wasn’t

footballs being thrown, but beams of light fired across the length of the pitch. The concentrated

streaks of light lasted for a few minutes, sent from an emitter on one side of the Jacksonville

Jaguars’ stadium and collected on a screen on the other.

The light had been collected from the Sun and then beamed out by large lenses on the field,

each about 1.2 metres (4ft) tall, acting like magnifying glasses. “We had to get up a ladder to

pull the cover off”, said Andrew Rush, chief executive of the Florida-based company Star Catcher,

which carried out the test. “We beamed 100 watts about 105 metres”.

The goal was simple: to see if sunlight could be beamed across space to power satellites or

even cities on Earth.

Space-based solar power aims to capture sunlight in orbit and _______________ it to the

ground as clean, renewable energy. Solar panels on Earth are limited by the atmosphere, the

weather and the day-night cycle. But in space, sunlight can be collected almost constantly, at

much higher _______________.

He recalled telling his father about the idea, who thought it sounded rather unrealistic. Yet,

as engineer David Homfray explained, “Space-based solar power makes the energy transition

work”. Some estimates suggest it could one day provide up to 80% of Europe’s renewable energy

needs.

Still, making it work is no small task. It would require enormous satellite structures,

complex operations and countless rocket launches. And there are cheaper renewable solutions

that could come online far faster — a crucial point if the world is to replace fossil fuels in time to

limit global warming.

The concept itself is not new. Science-fiction writer Isaac Asimov imagined it in 1941, and

NASA studied it in the 1970s and 1990s. Back then, it was dismissed as too costly and technically

unrealistic. But modern advances in rocket design, robotics and energy transmission have made

it more feasible.

Today, several countries — including the US, the UK, Japan and China — are investing in

prototypes. One approach involves using infrared lasers to beam power to the ground, while

another envisions vast solar stations in orbit that could supply electricity to entire countries.

However, not everyone agrees it’s viable. Critics warn about the risks of increasing space

_______________ and the difficulty of managing thousands of satellites safely. Others question

whether such huge orbital structures would even be legal under international space treaties.

NASA’s recent reports say the technology remains more expensive than terrestrial

renewables, yet progress is undeniable. “Once you make that initial investment”, said Michelle

Hanlon, a space-law expert, “that power is literally free”.

Whether space-based solar power will ever become practical remains uncertain. But for

now, it represents a bold vision of how humanity might one day capture sunlight beyond the

atmosphere — and beam power directly from the stars.

(Available at: www.bbc.com/future/article/20251029-the-beam-dream-should-we-build-solar-farms-in-space– text specially adapted for this test).

Analyse the following statements about some grammatical structures in the text:

I. The verb form “could finally make” (l. 02) expresses a future possibility.
II. The sentence “The light had been collected from the Sun” (l. 07) is in the passive voice.
III. The clause “whether such huge orbital structures would even be legal” (l. 34) expresses a condition.

Which ones are correct?

 

Provas

Questão presente nas seguintes provas
Read the following text and answer the questions.
Artificial Intelligence: The “lethal trifecta”
    LARGE LANGUAGE MODELS (LLMs), a trendy way of building artificial intelligence, have an inherent security problem: they cannot separate code from data. As a result, they are at risk of a type of attack called a prompt injection, in which they are tricked into following commands they should not. Sometimes the result is merely embarrassing, as when a customer-help agent is persuaded to talk like a pirate. On other occasions, it is far more damaging.
    The worst effects of this flaw are reserved for those who create what is known as the “lethal trifecta”. If a company, eager to offer a powerful AI assistant to its employees, gives an LLM access to untrusted data, the ability to read valuable secrets and the ability to communicate with the outside world at the same time, then trouble is sure to follow. And avoiding this is not just a matter for AI engineers. Ordinary users, too, need to learn how to use AI safely, because installing the wrong combination of apps can generate the trifecta accidentally. 
   Better AI engineering is, though, the first line of defence. And that means AI engineers need to start thinking like engineers, who build things like bridges and therefore know that shoddy work costs lives.
  The great works of Victorian England were erected by engineers who could not be sure of the properties of the materials they were using. In particular, whether by incompetence or malfeasance, the iron of the period was often not up to snuff. As a consequence, engineers erred on the side of caution, overbuilding to incorporate redundancy into their creations. The result was a series of centuries-spanning masterpieces.
   AI-security providers do not think like this. Conventional coding is a deterministic practice. Security vulnerabilities are seen as errors to be fixed, and when fixed, they go away. AI engineers, inculcated in this way of thinking from their schooldays, therefore often act as if problems can be solved just with more training data and more astute system prompts.
   These do, indeed, reduce risk. The cleverest frontier models are better at spotting and refusing malicious requests than their older or smaller cousins. But they cannot eliminate risk altogether. Unlike most software, LLMs are probabilistic. Their output is driven by random selection from likely responses. A deterministic approach to safety is thus inadequate. A better way forward is to copy engineers in the physical world and learn to work with, rather than against, capricious systems that can never be guaranteed to function as they should. That means becoming happier dealing with unpredictability by introducing safety margins, risk tolerance and error rates.
   Overbuilding in the AI age might, for instance, mean using a more powerful model than is needed for the task at hand, to reduce the risk it will be tricked into doing something inappropriate. It might mean imposing limits on the number of queries LLMs can take from external sources, calibrated to the risk of damage from a malicious query. And mechanical engineering emphasises failing safely. If an AI system must have access to secrets, then avoid handing it the keys to the kingdom.
   In the physical world, bridges have weight limits – even if they are not always stated clearly to drivers. And, importantly, these are well within the actual tolerances that calculations suggest a bridge will bear. The time has now come for the virtual world of AI systems to be similarly equipped.
Adapted from The Economist, September 27th, 2025, p. 10
The text concludes that the Victorian engineers’ decision
 

Provas

Questão presente nas seguintes provas
Read the following text and answer the questions.
Artificial Intelligence: The “lethal trifecta”
    LARGE LANGUAGE MODELS (LLMs), a trendy way of building artificial intelligence, have an inherent security problem: they cannot separate code from data. As a result, they are at risk of a type of attack called a prompt injection, in which they are tricked into following commands they should not. Sometimes the result is merely embarrassing, as when a customer-help agent is persuaded to talk like a pirate. On other occasions, it is far more damaging.
    The worst effects of this flaw are reserved for those who create what is known as the “lethal trifecta”. If a company, eager to offer a powerful AI assistant to its employees, gives an LLM access to untrusted data, the ability to read valuable secrets and the ability to communicate with the outside world at the same time, then trouble is sure to follow. And avoiding this is not just a matter for AI engineers. Ordinary users, too, need to learn how to use AI safely, because installing the wrong combination of apps can generate the trifecta accidentally. 
   Better AI engineering is, though, the first line of defence. And that means AI engineers need to start thinking like engineers, who build things like bridges and therefore know that shoddy work costs lives.
  The great works of Victorian England were erected by engineers who could not be sure of the properties of the materials they were using. In particular, whether by incompetence or malfeasance, the iron of the period was often not up to snuff. As a consequence, engineers erred on the side of caution, overbuilding to incorporate redundancy into their creations. The result was a series of centuries-spanning masterpieces.
   AI-security providers do not think like this. Conventional coding is a deterministic practice. Security vulnerabilities are seen as errors to be fixed, and when fixed, they go away. AI engineers, inculcated in this way of thinking from their schooldays, therefore often act as if problems can be solved just with more training data and more astute system prompts.
   These do, indeed, reduce risk. The cleverest frontier models are better at spotting and refusing malicious requests than their older or smaller cousins. But they cannot eliminate risk altogether. Unlike most software, LLMs are probabilistic. Their output is driven by random selection from likely responses. A deterministic approach to safety is thus inadequate. A better way forward is to copy engineers in the physical world and learn to work with, rather than against, capricious systems that can never be guaranteed to function as they should. That means becoming happier dealing with unpredictability by introducing safety margins, risk tolerance and error rates.
   Overbuilding in the AI age might, for instance, mean using a more powerful model than is needed for the task at hand, to reduce the risk it will be tricked into doing something inappropriate. It might mean imposing limits on the number of queries LLMs can take from external sources, calibrated to the risk of damage from a malicious query. And mechanical engineering emphasises failing safely. If an AI system must have access to secrets, then avoid handing it the keys to the kingdom.
   In the physical world, bridges have weight limits – even if they are not always stated clearly to drivers. And, importantly, these are well within the actual tolerances that calculations suggest a bridge will bear. The time has now come for the virtual world of AI systems to be similarly equipped.
Adapted from The Economist, September 27th, 2025, p. 10
The metaphor used in avoid handing it the keys to the kingdom (7th paragraph) means avoid giving the system
 

Provas

Questão presente nas seguintes provas
Read the following text and answer the questions.
Artificial Intelligence: The “lethal trifecta”
    LARGE LANGUAGE MODELS (LLMs), a trendy way of building artificial intelligence, have an inherent security problem: they cannot separate code from data. As a result, they are at risk of a type of attack called a prompt injection, in which they are tricked into following commands they should not. Sometimes the result is merely embarrassing, as when a customer-help agent is persuaded to talk like a pirate. On other occasions, it is far more damaging.
    The worst effects of this flaw are reserved for those who create what is known as the “lethal trifecta”. If a company, eager to offer a powerful AI assistant to its employees, gives an LLM access to untrusted data, the ability to read valuable secrets and the ability to communicate with the outside world at the same time, then trouble is sure to follow. And avoiding this is not just a matter for AI engineers. Ordinary users, too, need to learn how to use AI safely, because installing the wrong combination of apps can generate the trifecta accidentally. 
   Better AI engineering is, though, the first line of defence. And that means AI engineers need to start thinking like engineers, who build things like bridges and therefore know that shoddy work costs lives.
  The great works of Victorian England were erected by engineers who could not be sure of the properties of the materials they were using. In particular, whether by incompetence or malfeasance, the iron of the period was often not up to snuff. As a consequence, engineers erred on the side of caution, overbuilding to incorporate redundancy into their creations. The result was a series of centuries-spanning masterpieces.
   AI-security providers do not think like this. Conventional coding is a deterministic practice. Security vulnerabilities are seen as errors to be fixed, and when fixed, they go away. AI engineers, inculcated in this way of thinking from their schooldays, therefore often act as if problems can be solved just with more training data and more astute system prompts.
   These do, indeed, reduce risk. The cleverest frontier models are better at spotting and refusing malicious requests than their older or smaller cousins. But they cannot eliminate risk altogether. Unlike most software, LLMs are probabilistic. Their output is driven by random selection from likely responses. A deterministic approach to safety is thus inadequate. A better way forward is to copy engineers in the physical world and learn to work with, rather than against, capricious systems that can never be guaranteed to function as they should. That means becoming happier dealing with unpredictability by introducing safety margins, risk tolerance and error rates.
   Overbuilding in the AI age might, for instance, mean using a more powerful model than is needed for the task at hand, to reduce the risk it will be tricked into doing something inappropriate. It might mean imposing limits on the number of queries LLMs can take from external sources, calibrated to the risk of damage from a malicious query. And mechanical engineering emphasises failing safely. If an AI system must have access to secrets, then avoid handing it the keys to the kingdom.
   In the physical world, bridges have weight limits – even if they are not always stated clearly to drivers. And, importantly, these are well within the actual tolerances that calculations suggest a bridge will bear. The time has now come for the virtual world of AI systems to be similarly equipped.
Adapted from The Economist, September 27th, 2025, p. 10
Introducing in by introducing safety margins (6th paragraph) is similar in meaning to
 

Provas

Questão presente nas seguintes provas
Read the following text and answer the questions.
Artificial Intelligence: The “lethal trifecta”
    LARGE LANGUAGE MODELS (LLMs), a trendy way of building artificial intelligence, have an inherent security problem: they cannot separate code from data. As a result, they are at risk of a type of attack called a prompt injection, in which they are tricked into following commands they should not. Sometimes the result is merely embarrassing, as when a customer-help agent is persuaded to talk like a pirate. On other occasions, it is far more damaging.
    The worst effects of this flaw are reserved for those who create what is known as the “lethal trifecta”. If a company, eager to offer a powerful AI assistant to its employees, gives an LLM access to untrusted data, the ability to read valuable secrets and the ability to communicate with the outside world at the same time, then trouble is sure to follow. And avoiding this is not just a matter for AI engineers. Ordinary users, too, need to learn how to use AI safely, because installing the wrong combination of apps can generate the trifecta accidentally. 
   Better AI engineering is, though, the first line of defence. And that means AI engineers need to start thinking like engineers, who build things like bridges and therefore know that shoddy work costs lives.
  The great works of Victorian England were erected by engineers who could not be sure of the properties of the materials they were using. In particular, whether by incompetence or malfeasance, the iron of the period was often not up to snuff. As a consequence, engineers erred on the side of caution, overbuilding to incorporate redundancy into their creations. The result was a series of centuries-spanning masterpieces.
   AI-security providers do not think like this. Conventional coding is a deterministic practice. Security vulnerabilities are seen as errors to be fixed, and when fixed, they go away. AI engineers, inculcated in this way of thinking from their schooldays, therefore often act as if problems can be solved just with more training data and more astute system prompts.
   These do, indeed, reduce risk. The cleverest frontier models are better at spotting and refusing malicious requests than their older or smaller cousins. But they cannot eliminate risk altogether. Unlike most software, LLMs are probabilistic. Their output is driven by random selection from likely responses. A deterministic approach to safety is thus inadequate. A better way forward is to copy engineers in the physical world and learn to work with, rather than against, capricious systems that can never be guaranteed to function as they should. That means becoming happier dealing with unpredictability by introducing safety margins, risk tolerance and error rates.
   Overbuilding in the AI age might, for instance, mean using a more powerful model than is needed for the task at hand, to reduce the risk it will be tricked into doing something inappropriate. It might mean imposing limits on the number of queries LLMs can take from external sources, calibrated to the risk of damage from a malicious query. And mechanical engineering emphasises failing safely. If an AI system must have access to secrets, then avoid handing it the keys to the kingdom.
   In the physical world, bridges have weight limits – even if they are not always stated clearly to drivers. And, importantly, these are well within the actual tolerances that calculations suggest a bridge will bear. The time has now come for the virtual world of AI systems to be similarly equipped.
Adapted from The Economist, September 27th, 2025, p. 10
The phrase shoddy work costs lives (3rd paragraph) refers to work that is
 

Provas

Questão presente nas seguintes provas
Read the following text and answer the questions.
Artificial Intelligence: The “lethal trifecta”
    LARGE LANGUAGE MODELS (LLMs), a trendy way of building artificial intelligence, have an inherent security problem: they cannot separate code from data. As a result, they are at risk of a type of attack called a prompt injection, in which they are tricked into following commands they should not. Sometimes the result is merely embarrassing, as when a customer-help agent is persuaded to talk like a pirate. On other occasions, it is far more damaging.
    The worst effects of this flaw are reserved for those who create what is known as the “lethal trifecta”. If a company, eager to offer a powerful AI assistant to its employees, gives an LLM access to untrusted data, the ability to read valuable secrets and the ability to communicate with the outside world at the same time, then trouble is sure to follow. And avoiding this is not just a matter for AI engineers. Ordinary users, too, need to learn how to use AI safely, because installing the wrong combination of apps can generate the trifecta accidentally. 
   Better AI engineering is, though, the first line of defence. And that means AI engineers need to start thinking like engineers, who build things like bridges and therefore know that shoddy work costs lives.
  The great works of Victorian England were erected by engineers who could not be sure of the properties of the materials they were using. In particular, whether by incompetence or malfeasance, the iron of the period was often not up to snuff. As a consequence, engineers erred on the side of caution, overbuilding to incorporate redundancy into their creations. The result was a series of centuries-spanning masterpieces.
   AI-security providers do not think like this. Conventional coding is a deterministic practice. Security vulnerabilities are seen as errors to be fixed, and when fixed, they go away. AI engineers, inculcated in this way of thinking from their schooldays, therefore often act as if problems can be solved just with more training data and more astute system prompts.
   These do, indeed, reduce risk. The cleverest frontier models are better at spotting and refusing malicious requests than their older or smaller cousins. But they cannot eliminate risk altogether. Unlike most software, LLMs are probabilistic. Their output is driven by random selection from likely responses. A deterministic approach to safety is thus inadequate. A better way forward is to copy engineers in the physical world and learn to work with, rather than against, capricious systems that can never be guaranteed to function as they should. That means becoming happier dealing with unpredictability by introducing safety margins, risk tolerance and error rates.
   Overbuilding in the AI age might, for instance, mean using a more powerful model than is needed for the task at hand, to reduce the risk it will be tricked into doing something inappropriate. It might mean imposing limits on the number of queries LLMs can take from external sources, calibrated to the risk of damage from a malicious query. And mechanical engineering emphasises failing safely. If an AI system must have access to secrets, then avoid handing it the keys to the kingdom.
   In the physical world, bridges have weight limits – even if they are not always stated clearly to drivers. And, importantly, these are well within the actual tolerances that calculations suggest a bridge will bear. The time has now come for the virtual world of AI systems to be similarly equipped.
Adapted from The Economist, September 27th, 2025, p. 10
The word tricked (1st paragraph) means that LLMs can be
 

Provas

Questão presente nas seguintes provas
Read the following text and answer the questions.
Artificial Intelligence: The “lethal trifecta”
    LARGE LANGUAGE MODELS (LLMs), a trendy way of building artificial intelligence, have an inherent security problem: they cannot separate code from data. As a result, they are at risk of a type of attack called a prompt injection, in which they are tricked into following commands they should not. Sometimes the result is merely embarrassing, as when a customer-help agent is persuaded to talk like a pirate. On other occasions, it is far more damaging.
    The worst effects of this flaw are reserved for those who create what is known as the “lethal trifecta”. If a company, eager to offer a powerful AI assistant to its employees, gives an LLM access to untrusted data, the ability to read valuable secrets and the ability to communicate with the outside world at the same time, then trouble is sure to follow. And avoiding this is not just a matter for AI engineers. Ordinary users, too, need to learn how to use AI safely, because installing the wrong combination of apps can generate the trifecta accidentally. 
   Better AI engineering is, though, the first line of defence. And that means AI engineers need to start thinking like engineers, who build things like bridges and therefore know that shoddy work costs lives.
  The great works of Victorian England were erected by engineers who could not be sure of the properties of the materials they were using. In particular, whether by incompetence or malfeasance, the iron of the period was often not up to snuff. As a consequence, engineers erred on the side of caution, overbuilding to incorporate redundancy into their creations. The result was a series of centuries-spanning masterpieces.
   AI-security providers do not think like this. Conventional coding is a deterministic practice. Security vulnerabilities are seen as errors to be fixed, and when fixed, they go away. AI engineers, inculcated in this way of thinking from their schooldays, therefore often act as if problems can be solved just with more training data and more astute system prompts.
   These do, indeed, reduce risk. The cleverest frontier models are better at spotting and refusing malicious requests than their older or smaller cousins. But they cannot eliminate risk altogether. Unlike most software, LLMs are probabilistic. Their output is driven by random selection from likely responses. A deterministic approach to safety is thus inadequate. A better way forward is to copy engineers in the physical world and learn to work with, rather than against, capricious systems that can never be guaranteed to function as they should. That means becoming happier dealing with unpredictability by introducing safety margins, risk tolerance and error rates.
   Overbuilding in the AI age might, for instance, mean using a more powerful model than is needed for the task at hand, to reduce the risk it will be tricked into doing something inappropriate. It might mean imposing limits on the number of queries LLMs can take from external sources, calibrated to the risk of damage from a malicious query. And mechanical engineering emphasises failing safely. If an AI system must have access to secrets, then avoid handing it the keys to the kingdom.
   In the physical world, bridges have weight limits – even if they are not always stated clearly to drivers. And, importantly, these are well within the actual tolerances that calculations suggest a bridge will bear. The time has now come for the virtual world of AI systems to be similarly equipped.
Adapted from The Economist, September 27th, 2025, p. 10
By referring to LLMs as a trendy way of building artificial intelligence (1st paragraph), the author implies they are
 

Provas

Questão presente nas seguintes provas
Read the following text and answer the questions.
Artificial Intelligence: The “lethal trifecta”
    LARGE LANGUAGE MODELS (LLMs), a trendy way of building artificial intelligence, have an inherent security problem: they cannot separate code from data. As a result, they are at risk of a type of attack called a prompt injection, in which they are tricked into following commands they should not. Sometimes the result is merely embarrassing, as when a customer-help agent is persuaded to talk like a pirate. On other occasions, it is far more damaging.
    The worst effects of this flaw are reserved for those who create what is known as the “lethal trifecta”. If a company, eager to offer a powerful AI assistant to its employees, gives an LLM access to untrusted data, the ability to read valuable secrets and the ability to communicate with the outside world at the same time, then trouble is sure to follow. And avoiding this is not just a matter for AI engineers. Ordinary users, too, need to learn how to use AI safely, because installing the wrong combination of apps can generate the trifecta accidentally. 
   Better AI engineering is, though, the first line of defence. And that means AI engineers need to start thinking like engineers, who build things like bridges and therefore know that shoddy work costs lives.
  The great works of Victorian England were erected by engineers who could not be sure of the properties of the materials they were using. In particular, whether by incompetence or malfeasance, the iron of the period was often not up to snuff. As a consequence, engineers erred on the side of caution, overbuilding to incorporate redundancy into their creations. The result was a series of centuries-spanning masterpieces.
   AI-security providers do not think like this. Conventional coding is a deterministic practice. Security vulnerabilities are seen as errors to be fixed, and when fixed, they go away. AI engineers, inculcated in this way of thinking from their schooldays, therefore often act as if problems can be solved just with more training data and more astute system prompts.
   These do, indeed, reduce risk. The cleverest frontier models are better at spotting and refusing malicious requests than their older or smaller cousins. But they cannot eliminate risk altogether. Unlike most software, LLMs are probabilistic. Their output is driven by random selection from likely responses. A deterministic approach to safety is thus inadequate. A better way forward is to copy engineers in the physical world and learn to work with, rather than against, capricious systems that can never be guaranteed to function as they should. That means becoming happier dealing with unpredictability by introducing safety margins, risk tolerance and error rates.
   Overbuilding in the AI age might, for instance, mean using a more powerful model than is needed for the task at hand, to reduce the risk it will be tricked into doing something inappropriate. It might mean imposing limits on the number of queries LLMs can take from external sources, calibrated to the risk of damage from a malicious query. And mechanical engineering emphasises failing safely. If an AI system must have access to secrets, then avoid handing it the keys to the kingdom.
   In the physical world, bridges have weight limits – even if they are not always stated clearly to drivers. And, importantly, these are well within the actual tolerances that calculations suggest a bridge will bear. The time has now come for the virtual world of AI systems to be similarly equipped.
Adapted from The Economist, September 27th, 2025, p. 10
The author compares AI and 19th century engineers to argue that the latter were
 

Provas

Questão presente nas seguintes provas
Read the following text and answer the questions.
Artificial Intelligence: The “lethal trifecta”
    LARGE LANGUAGE MODELS (LLMs), a trendy way of building artificial intelligence, have an inherent security problem: they cannot separate code from data. As a result, they are at risk of a type of attack called a prompt injection, in which they are tricked into following commands they should not. Sometimes the result is merely embarrassing, as when a customer-help agent is persuaded to talk like a pirate. On other occasions, it is far more damaging.
    The worst effects of this flaw are reserved for those who create what is known as the “lethal trifecta”. If a company, eager to offer a powerful AI assistant to its employees, gives an LLM access to untrusted data, the ability to read valuable secrets and the ability to communicate with the outside world at the same time, then trouble is sure to follow. And avoiding this is not just a matter for AI engineers. Ordinary users, too, need to learn how to use AI safely, because installing the wrong combination of apps can generate the trifecta accidentally. 
   Better AI engineering is, though, the first line of defence. And that means AI engineers need to start thinking like engineers, who build things like bridges and therefore know that shoddy work costs lives.
  The great works of Victorian England were erected by engineers who could not be sure of the properties of the materials they were using. In particular, whether by incompetence or malfeasance, the iron of the period was often not up to snuff. As a consequence, engineers erred on the side of caution, overbuilding to incorporate redundancy into their creations. The result was a series of centuries-spanning masterpieces.
   AI-security providers do not think like this. Conventional coding is a deterministic practice. Security vulnerabilities are seen as errors to be fixed, and when fixed, they go away. AI engineers, inculcated in this way of thinking from their schooldays, therefore often act as if problems can be solved just with more training data and more astute system prompts.
   These do, indeed, reduce risk. The cleverest frontier models are better at spotting and refusing malicious requests than their older or smaller cousins. But they cannot eliminate risk altogether. Unlike most software, LLMs are probabilistic. Their output is driven by random selection from likely responses. A deterministic approach to safety is thus inadequate. A better way forward is to copy engineers in the physical world and learn to work with, rather than against, capricious systems that can never be guaranteed to function as they should. That means becoming happier dealing with unpredictability by introducing safety margins, risk tolerance and error rates.
   Overbuilding in the AI age might, for instance, mean using a more powerful model than is needed for the task at hand, to reduce the risk it will be tricked into doing something inappropriate. It might mean imposing limits on the number of queries LLMs can take from external sources, calibrated to the risk of damage from a malicious query. And mechanical engineering emphasises failing safely. If an AI system must have access to secrets, then avoid handing it the keys to the kingdom.
   In the physical world, bridges have weight limits – even if they are not always stated clearly to drivers. And, importantly, these are well within the actual tolerances that calculations suggest a bridge will bear. The time has now come for the virtual world of AI systems to be similarly equipped.
Adapted from The Economist, September 27th, 2025, p. 10

Based on the text, mark the statements below as true (T) or false (F).

I. AI models are watertight when it comes to safety risks.

II Bridges built in the Victorian Age were proven to be quite fragile.

III. A deterministic model does not deal with randomness.

The statements are, respectively,

 

Provas

Questão presente nas seguintes provas