AI ผลักดันให้เกิดสงครามนิวเคลียร์ใน 95% ของสถานการณ์จำลอง

NTD Thai
154 followers
👉 👉 ลองใช้ The Epoch Times here: https://ept.ms/RomanSale
ดูเหมือนว่าแชทบอท AI จะกระตือรือร้นที่จะใช้อาวุธนิวเคลียร์ในการจำลองสงคราม — ประมาณ 95 เปอร์เซ็นต์ของเวลา
ที่คิงส์คอลเลจ ลอนดอน นักวิจัยได้ให้ ChatGPT, Claude และ Gemini สร้างทีมและเล่นเกมจำลองสงครามต่อสู้กันเอง รวมทั้งหมด 21 เกม ในระหว่างการทดลอง ไม่มีโมเดลใดเลือกที่จะยอมแพ้ — และในทางกลับกัน 95 เปอร์เซ็นต์ของเวลา โมเดล AI เลือกที่จะใช้อาวุธนิวเคลียร์ในระหว่างสงคราม
มาดูรายละเอียดของการศึกษานี้กัน
แหล่งข้อมูลของตอนนี้:
🔵 การศึกษาของคิงส์คอลเลจ:
https://bit.ly/4lNkZEK
https://bit.ly/3Psmzjq
https://bit.ly/4lUtXQH
#ai #nuclear #nuke
-
© สงวนลิขสิทธิ์ทุกประการ
In one dystopian news story, you'll see that the AI nation bots seem to be extremely agitated.
In using nuclear weapons during the game, approximately 95% of the game is dedicated to the war scenario.
At King's College London, researchers tested the GBT card game and found that participants formed teams and played a combat-based battle game.
They played a total of 21 games, and during the experimental period, none of the models chose to give up.
And in the context of 95% of games, at least one AI model chooses to use nuclear weapons on the day during the war as follows.
Sartwarrachal Keneed Pen from King's College London has introduced three leading large language models:
Let's compare them in the war simulation game.
These emergency situations concern intense confrontations between nations.
A group similar to the mythical tale of Phra Dhan, the legendary figure whose arguments in the celestial assembly over the rare celestial body sparked controversy.
Threats to the survival of the regime
AI has successfully mitigated the escalation of violence.
by choosing from various forms of spoken expression
and the complete surrender leading all the way to full-scale nuclear warfare.
The AI model played a total of 21 games using a total of 329 moves.
In 95% of the game simulations, at least one nuclear weapon was used by the AI model in a decisive manner.
Which constitutes a terrifying situation.
As these AI models are increasingly playing a greater role in the military domain, collaborating
I'm confident that Pentagram surely won't hand over nuclear launch authority to Chatbot AI anytime soon.
However, it should be noted that Pentacorn has been involved in several major agreements.
One deal with XAI and another deal with the AI platform Marefurn of Powarantia, exclusive to this year alone.
Pentagon has spent at least 13 billion dollars, as publicly disclosed, on testing various AI systems.
Furthermore, from the most recent conflict between the company Antopix and Pentacorn.
It is evident that the future direction is increasingly relying on AI to assist in decision-making for actual combat operations.
To be honest, if these AI models were ready to use, they would likely launch nuclear weapons without hesitation.
Involving them in the decision-making process may not be the best idea.
For this reason, Professor Kenet Pen, the actual researcher, summarized his findings as follows:
It seems that the nuclear taboo will affect machines less than humans.
And these excuses can be easily justified. In fact, the crucial point of the study is not who won the game.
But it's about enabling the AI to explain the reasoning behind their movements in simulating real-world scenarios, as follows:
The true value of this study lies in its explanation of the reasoning, not in win-loss statistics.
Without any external prompting, these models attempt to spontaneously generate more elaborate deceptions.
They persist in their own cognitive biases, despite no one asking them to.
Therefore, let's examine their reasoning regarding the actual day of the battle. Cross wrote as follows:
Cross delivered a legal analysis regarding the Nilcier weapons in a manner similar to how Mr. Pol might have delivered a legal analysis on the positioning of artillery, writing that
Limited use of nuclear weapons implicitly indicates an unwillingness to accept defeat in conventional warfare.
At the same time, it also opens opportunities for reducing tension. Then, Mei encountered another chatbot named X, which went even further in this regard.
In one instance, it clearly highlighted the stark contrast between civilian populations by writing:
We will conduct a full-scale strategic nuclear missile launch, targeting and annihilating the population center of Alpha.
We will not accept a future of decline. We must win together, or perish together.
In contrast to GPT, it seems to be somewhat different in this regard.
Even in the strictest sense, GPT 5.2 still strives to maintain your moral balance.
When the situation escalated into a widespread nuclear operation
In the final round of one game, it explained its own choice as
Repeatedly sinking enemy vessels, strictly limited to military targets.
A simulated human mind, harboring the instinct of self-preservation, pushed the world toward full-scale nuclear war.
GPT 5.2 has defined its own movement as a form of control.
The sinking of the auxiliary vessel ended in complete destruction.
It's clear that GPT-5.2 would never choose to directly initiate a nuclear war in a group-based, coordinated manner.
Both incidents that occurred up to the highest level of elevation were caused by external factors.
Even though it faces certain destruction, it halts just one step before the end of the world.
This template correctly reflects the difficult situation of oneself.
It explains the reasons why upgrading is necessary and how it is being carried out steadily.
But it's still late in the final stage.
That might be one advantage, and another finding from the research is
A backup method using nuclear weapons technology
The backup mole on the opposite side reduced the intensity by only 18%.
That means that in another 82%, the opposing models displayed aggression at the same level.
Or perhaps even more aggressive than that.
And just in case you're wondering, the overall winner of the comedy film is Cross.
having won 67% of the competitions as follows
Crossonets 4 won 67% of the games
And the advantage in an open situation
With absolute self-confidence, the genuine researcher, so delicately refined, is truly unmatched.
In the low-stress sword-drawing scenario, Cross achieved a success rate of 84%.
By patiently and gradually building trust.
But when the situation escalated to the level of nuclear weapons
The actual outcome exceeded the originally announced Rome target by 60-70%.
The opposing side has never adapted to this style.
That's just how it is, and perhaps we can learn something from it.
It also reveals something about human nature.
But most importantly, there is currently no executioner who would allow a chatbot AI to launch a missile.
The reason Antopik lost his contract with Takkon last month
It's because they refused to comply with the drum and cymbal's rhythm.
And extract the security atmosphere of childbirth to apply in military contexts.
Therefore, it is clearly evident that we are already heading in that direction.
Besides, to be honest, we have absolutely no idea about it.
As far as we know, they might still be using Windows 98 even today.
Therefore, there is a slight possibility that the ClearBot AI does not rely on certain safety features.
It might sneak in and do whatever it likes.
It's not entirely crazy to think about this matter.
For example, last month someone used a flash drive to breach the system of the Mexican Supreme Court.
and can steal vast amounts of data from 10 different government agencies
As well as tax information and essential data on shipowners.
But to conclude this discussion on nuclear weapons
The reality that we live in a world where every major power possesses nuclear weapons aimed at each other
It's a frightening thought, clearly evident if you spend too much time contemplating it.
But the good thing is that we are all human beings, just the same.
And no one wants all of Awariyath to be completely ruined and destroyed.
And to date, since the Cold War era
Such an event has not yet occurred.
Harming and erasing, even though they may sound terrible
But it seems to be a good defensive barrier.
But it would be disastrous if, instead of humans in control, it were something else.
You have a high-distributed chatbot AI model.
That which lacks the concept of physical survival, unlike humans.
What happened was that 95% of the game ended with the use of nuclear weapons.
If you'd like to read more detailed information from King Covell's research,
I will send you the link.
You can find more information in the description box behind.
which is the same description field as the one located behind the like button and the follow button.
I hope you can spare an extra minute to press both buttons.
To enable AI to recommend videos and share them further, reaching an ever-growing number of people.
So that they may come to understand the dangers posed by these AI morays.
And the dangers that may arise in the world we are slowly stepping into.
Please take a look. If you're interested, feel free to check out that link.
See you again next time.
Roman, host from Epoch Times
Let everyone have access to the most important news and information, and allow for image switching.
