Skip to main content

Why AI writes something that seems like bullshit?

Image: Pinterest


Some people are worrying if AI is making things like the thesis. The fact is that AI can make things that seem very academic and qualified. But the text itself has no sense. Why computer cannot make that thing like humans? 

If we want to make an AI that can find information from the Internet we must select some method that this system uses for selecting the texts that it connects to the document. Using linear programming methodology is almost impossible to create text that is suitable for anything. 

In that model, the system selects a group of texts. And then it starts to connect them to one entirety. The system can put those texts in cerrtain order by putting their hashtags in alphabetical order. 


And then the system can start its operation. The first paragraph is taken from the first text. 


The second one is taken from the second text. And by using this method the system will make nice-looking ten-paragraph text. 

The problem is that the system selects those texts that it connects by using a certain methodology. And one of those methods is that the system selects ten of the top-ranked pages from Google. Or the system can use only hashtags for sorting those pages. 

Because the program itself will not understand what those pages involved, the result might be something else than we want. This is the limit of AI. The AI that understands the text is almost impossible to make. 

We can make the AI that answers our questions and looks very reasonable. When we ask AI what it can do, it can give impressive answers. The AI can say that It's AI and know many things about things like astronomy. The thing that makes AI look very wise is that it has a database of the answers to the most common astronomical questions.

The system might involve details of the most common stars their spectral class etc. It can have information about the birth of the stars, and if somebody asks something about them it simply drives that data to the loudspeaker.


Those databases are like tapes. They involve information that is needed for the answers about astronomy. 

  

 But if somebody asks a question about history or politics, the AI can smply answer that those things belong to some other AI, and the expertise of that AI is astronomy. Then that tape can involve the mention that the AI should stop quite soon because there is limited time. 

That thing seems very complicated. But it's a large group of sound files. When somebody asks the question. Where are words that are linking it with things like Jupiter moons computer will find the Jupiter database? Then the system can ask, would that person want to hear something about all moons? 

Or if a person answers that Galilean moons are interesting. The command activates the Galilean moon databases. 

And the computer can read all three databases one after one. That means the AI might look very smart. But it's nothing more than a series of sound files that are activated by using certain words. 


https://bigthink.com/the-present/ai-language-models-gpt-3-pseudo-profound-bullshit/. 


Labels:

Comments

Post a Comment

Popular posts from this blog

The LK-99 could be a fundamental advance even if it cannot reach superconductivity in 400K.

The next step in superconducting research is that LK-99 was not superconducting at room temperature. Or was it? The thing is that there is needed more research about that material. And even if it couldn't reach superconductivity in 400K that doesn't mean that material is not fundamental. And if LK-99 can maintain its superconductivity in 400K that means a fundamental breakthrough in superconducting technology.  The LK-99 can be hype or it can be the real thing. The thing is, anyway, that high-voltage cables and our electric networks are not turning superconducting before next summer. But if we can change the electric network to superconducting by using some reasonable material. That thing can be the next step in the environment. Superconductors decrease the need to produce electricity. But today cooling systems that need lots of energy are the thing that turn superconductors that need low temperatures non-practical for everyday use.  When the project begins there is lots of ent
Post a Comment
Read more

Black holes, the speed of light, and gravitational background are things that are connecting the universe.

 Black holes, the speed of light, and gravitational background are things that are connecting the universe.  Black holes and gravitational waves: is black hole's singularity at so high energy level that energy travels in one direction in the form of a gravitational wave.  We normally say that black holes do not send radiation. And we are wrong. Black holes send gravitational waves. Gravitational waves are wave movement or radiation. And that means the black holes are bright gravitational objects.  If we can use water to illustrate the gravitational interaction we can say that gravitational waves push the surface tension out from the gravitational center. Then the other quantum fields push particles or objects into a black hole. The gravitational waves push energy out from the objects. And then the energy or quantum fields behind that object push them into the gravitational center.  The elementary particles are quantum fields or whisk-looking structures. If the gravitational wave is
Post a Comment
Read more

The CEO of Open AI, Sam Altman said that AI development requires a similar organization as IAEA.

We know that there are many risks in AI development. And there must be something that puts people realize that these kinds of things are not jokes. The problem is how to take control of the AI development. If we think about international contracts regarding AI development. We must realize that there is a possibility that the contract that should limit AI development turns into another version of the Nuclear Non-Proliferation Treaty. That treaty didn't ever deny the escalation of nuclear weapons. And there is a big possibility that the AI-limitation contracts follow the route of the Nuclear Non-Proliferation Treaty.  The biggest problem with AI development is the new platforms that can run every complicated and effective code. That means the quantum computer-based neural networks can turn themselves more intelligent than humans. The AI has the ultimate ability to learn new things. And if it runs on the quantum-hybrid system that switches its state between binary and quantum states,
Post a Comment
Read more