Have you ever asked the artificial intelligence model to tell you a joke and receive something painless? Something like:
Why do programmers prefer the dark situation? Because light attracts insects!
Pioneer. truly.
Most of the time, these alleged jokes resulting from artificial intelligence are very strange and predictable. the reason? It’s a machine! They do not have a human context – all that he sees is patterns of the data they were trained in, and when we ask them to say a joke, they enrich the words that depend on the possibilities rather than actual intelligence or comedy timing.
This is not a missile science for a person working with artificial intelligence, but it is interesting to see What makes it as it is. With this understanding, you can also know how to do so Ask artificial intelligence models for better jokes.
Amnesty International “Learn” how it looks joke
Artificial Intelligence model (for example, GPT or a link of a joke) is not born in a sense of humor. Instead, he learns from thousands of examples of a broken joke from books, websites and conversations. But instead of to understand Humor as we do, it recognizes statistical patterns.
Below is an excessive look at how to treat the artificial intelligence model of structures (a completely different symbol, just for your understanding):
import random
setup = ["Why did the chicken cross the road?", "What do you get when you mix AI with a comedian?", "Why don't programmers like nature?"]
punchline = ["To get to the other side!", "A neural network with bad timing!", "Too many bugs."]
print(random.choice(setup) + " " + random.choice(punchline))
# NOTE: This is oversimplified for easier understanding.
# Real AI models don't just randomly select from pre-written jokes.
# They use neural networks with billions of parameters to predict
# the most likely next words based on patterns in their training data.
See the problem? He gets to know pattern From a joke: First, the preparation comes, and the second comes the pollen lines; He does not understand why a funny joke.
Artificial intelligence predicts the following words, not the most entertaining things
The outputs of artificial intelligence models, whether or not the spirit of humorThe first is working on statistical possibility. Big language models such as GPT often work by statistically following games (words). This is a great thing to generate automatic and essential completion, but it is terrible for the original humor.
For example, if artificial intelligence generates the preparation: “Why did the chicken cross the road?”
It predicts the joint responses of the puncline. Since “access to the other side” is the statistically most statistical answer, it will create this often – although it is less funny.
from collections import Counter
jokes_dataset = ["To get to the other side!", "Because it was programmed to!", "To escape the AI overlords!"]
probabilities = Counter(jokes_dataset)
# chooses the most frequent joke
print(probabilities.most_common(1)[0][0])
For this reason they can be really predicted and repeated.
Artificial intelligence lacks context and global knowledge
Humor, unlike generating the usual text, is deeply related to human emotions, experiences and cultures. It is sad to say, it is the fine things that you do not understand in the first place. Even with the models trained on huge amounts of data, artificial intelligence fights when it comes to:
- Mockery: “O great, another joke created from artificial intelligence!”
- WordPlay & Puns: “AI enters a tape … and fails to test Torring.”
- Social awareness: “Why did artificial intelligence go to treatment? Because he could not address his feelings.”
Although artificial intelligence can identify common joke formats, it is not possible Get Why is something funny. As a result, they are trying to be funny.
attempts = ["I'm hilarious, trust me!", "Why did the algorithm break up? It lost its spark!", "What's an AI's favorite drink? Java."]
print(random.choice(attempts))
Then again, who blames? These are the trained machines on the data that are not obtained by linking. The last thing, Amnesty International’s engineers want to prosecute to train artificial intelligence models on data extracted from reserve comedians.
Fix the humor of artificial intelligence
For better jokes than artificial intelligence, you should be specific. Tell him the way you want (perhaps like the dark humor of Louis CK or Chandler Ping), and set the scene, and let him know the type of humor you follow. It will not make it 50 times better, but at least it will not be clear and harsh.
This will give Amnesty International a framework through which to work. Because at the present time, without context and understanding of human emotions, all that can do is guess the punching line – usually with the most predictable and robotic ways.
These are just 700 words, not a summary. We see you next time!
