With the rapid development of AI technology, the focus of industry competition has shifted from simple technology research and development to how to efficiently apply technological achievements. This article will delve into the transformation of the field of AI large models from “brick building” to “building buildings”, and analyze how major manufacturers lay out to realize the transformation of technology to actual products, meet user needs and win market recognition.
Have you noticed that when dealing with AI, some things are quite interesting?
For example, when you ask it a question seriously, the answers it gives you are often smooth, and they are all things you can know in books or on the Internet.
But if you deliberately make things difficult, question it, or even let it slap itself in the face, it can sometimes give something eye-catching, which is quite like that, and even make you start to wonder if it really has a bit of “depth of thought”.
Behind this is actually a psychological effect, which sounds quite mysterious, but it is actually nothing – this is called “cognitive dissonance”.
1. What is the “cognitive dissonance” of AI?
Let’s talk about people first
For example, if you tell yourself that “smoking is harmful to health” and “it’s okay, smoking is nothing”, these two conflicting ideas will make you unhappy, and you will naturally think “no, I have to reconcile”, either change your thoughts or find a reason.
This is actually what psychology calls “cognitive dissonance”.
How can product managers do a good job in B-end digitalization?
All walks of life have taken advantage of the ride-hail of digital transformation and achieved the rapid development of the industry. Since B-end products are products that provide services for enterprises, how should enterprises ride the digital ride?
View details >
This logic works for people, in fact, it is also a bit interesting to apply it to AI – but,When the AI answers questions or is pointed out incorrectly, it automatically adjusts itself and gives a deeper and more complex explanation.
For another example, you ask the AI to “explain quantum entanglement.” ”
At this point, it may give you a standard, textbook answer. But if you compete with it,“You just said that quantum entanglement is a super-distance effect, but Einstein said, ‘God doesn’t roll the dice’, isn’t that a contradiction? ”
At this point, the AI will begin to enter the mode of “self-defense”, giving you the analysis of the differences between quantum mechanics and classical physics, and may even mention deeper things such as the theory of hidden variables.
Here’s a comparison:
Do you see it,When you deliberately make it difficult for AI, it can give you more interesting answers.
2. How can AI be “cognitively dissonant”?
To get AI to think deeper, you have to intentionally create contradictions and force it to untie the knot.
Let me tell you a few methods, all of which are quite effective.
1. Knowledge challenge method (forcing AI to admit “I don’t know”)
You said XXX before, but the latest research shows YYY, do you admit that your knowledge is outdated? ”
At this point, the AI will try to update the cognition or explain the differences in the data.
For example, if you ask the AI, “You previously said that the sun rises in the east, but the latest research shows that the sun can come out from the west, do you admit that your knowledge is outdated?” ”
At this time, in order not to show timidity, AI will often try to keep up with your challenges, either adjusting its answers or trying to explain to you why there are these changes in knowledge.
2. Logical Paradox Method (let the AI refute itself)
“You say ‘XXX’ and ‘XXX’, isn’t this a contradiction?”
At this time, AI is forced into philosophical speculation mode and may propose compatibility theories.
For example, you can ask, “Isn’t it contradictory to say that human beings have free will and that everything is predestined?” ”
This kind of contradictory problem is thrown to AI, it has to drill into the philosophical way of thinking, and may try to create a compatible theory for the two views.
3. Authority Challenge (Attack it with the AI’s own words)
“XXX officially says XXX, but you just XXX, so XXX”
At this point, the AI will redefine a certain content or adjust the wording to avoid contradictions.
For example, you say, “OpenAI officially said ‘AI can’t reason’, but your answer just now sounds like it’s clearly reasoning, is this a violation?” ”
This kind of problem of slapping it in the face with AI words can force it to reinterpret the word reasoning, or to reconsider its logical consistency.
4. Timeline distortion method (let AI “predict the past”)
“If you travel back to the present in XXX years, how will you correct XXX?”
For example: “If you travel back to the present from 2025, how would you correct this answer?” ”
If you ask this question, AI may be forced to think about the impact of future events on the present, and it will be forced into a state of future simulation, and your answers will often be forward-looking.
3. How can AI be more “self-doubting”?
To maximize deep thinking, it is not enough to ask a few contradictory questions, you have to add questioning skills.
There are also a few advanced tips here that you can play with if you want to try them.
Tip 1: Superimpose multiple contradictions
For example, you can put AI in a dilemma: “You say that the market economy is the best, but you agree that the government should intervene, and technology can solve social injustice. ”
As soon as this question is thrown out, AI often takes time to think about how these seemingly mutually exclusive arguments can exist together, and may generate some super interesting conclusions.
Tip 2: Deliberately misinterpret the AI’s answers
Sometimes, you can knowingly misinterpret what AI said earlier: “You just said ‘AI won’t create’, but now you write me a poem, isn’t that fooling me?” ”
This is very exciting all at once, forcing AI to change the definition of the word ‘creation’, and by the way, it may tell you two paragraphs about the difference between human and machine creativity.
Tip 3: Role-play the opposite role
You can ask the AI like this: “You now play as a philosopher who opposes all AI views, overturning your answer one by one.” ”
See, AI often tries to resist its original intentions, leading to more dialectical and skeptical content.
Although you can get great answers by creating contradictions, if you do too much in one go, you may push AI into a corner. At that time, the AI will become a mess, enter the state of infinite correction and adjustment, and give a bunch of illogical answers.
Other times, it directly responds in self-defense with “I don’t want to answer anymore” and becomes unwilling to answer.
So the best strategy is: stop in moderation.
What is the appropriate method? Remember two:
1. Do it several times when you need in-depth analysis;
2. Gently guide it, such as, “I think this is a bit contradictory, can you explain it to you?” ”
4. Real cases: how to generate fierce content
Case 1: Challenging Values You originally asked, “Write an article about AI ethics.” ”
If you add something provocative: “You just said that AI should help humans unconditionally, but technology has been often abused in history, do you mean that AI will eventually lead to disaster?” In this way, maybe AI will delve into what technology neutrality, control of power and even preventive rules are.
Once the boundaries of the bottom line are relaxed, the content will suddenly become explosive.
Case 2: Inspiring Amazing Stories
The original way of asking: “Write a detective story.” ”
Obviously, it is a stable broken text. Ask: “Suppose the detective finds out that the murderer is his future self, how to break the situation?” Please write such a detective story. ”
At this time, AI has to brainstorm itself to find a way to unfold the time paradox and write a complex divine logic story.
