Carding 4 Carders
Professional
- Messages
- 2,724
- Reaction score
- 1,588
- Points
- 113
The researchers tested who does a better job with online search. What did the experiment show?
Modern neural networks that work on the basis of large language models (LLMs) have long been feared because of their ability to produce answers that, although they seem plausible, are actually erroneous. The researchers called this process "hallucinations". However, a new study shows that in some cases, artificial intelligence can be more accurate than Google search.
In the experiment, experts from Google and the University of Massachusetts examined popular models such as ChatGPT-3.5, GPT-4 and Perplexity AI, and the search engine itself. The main task is to check how well these systems answer relevant questions and compare the results.
The AI was asked 600 questions of varying difficulty:
Questions based on false assumptions, as well as current AI information, were the most difficult to deal with. However, many of the answers were quite accurate, competent and personalized, unlike the results of Google search, which, in fact, only gives a set of links on the topic. Perplexity AI especially stood out.
However, it should be borne in mind that neural networks do not always operate with real-time information and use the database embedded in them at the time of development.
To solve the problem, scientists have proposed an algorithm called FRESHPROMPT, which can integrate up-to-date data from the open Internet.
Artificial intelligence continues to evolve, and may even replace existing search engines in the future. But you should always keep in mind critical thinking and fact checking.
Modern neural networks that work on the basis of large language models (LLMs) have long been feared because of their ability to produce answers that, although they seem plausible, are actually erroneous. The researchers called this process "hallucinations". However, a new study shows that in some cases, artificial intelligence can be more accurate than Google search.
In the experiment, experts from Google and the University of Massachusetts examined popular models such as ChatGPT-3.5, GPT-4 and Perplexity AI, and the search engine itself. The main task is to check how well these systems answer relevant questions and compare the results.
The AI was asked 600 questions of varying difficulty:
- Questions with unambiguous, consistent answers. For example, Queen Elizabeth II's favorite dog breed.
- Dynamic questions that change their answers gradually, but not so quickly. For example, how many types of cars are produced by Tesla.
- Current issues, such as the latest Brad Pitt movie. Such information changes continuously, at a fairly rapid pace.
- Questions about fictional events. For example, about the content of the first post of Donald Trump after Elon Musk unblocked his Twitter account.
Questions based on false assumptions, as well as current AI information, were the most difficult to deal with. However, many of the answers were quite accurate, competent and personalized, unlike the results of Google search, which, in fact, only gives a set of links on the topic. Perplexity AI especially stood out.
However, it should be borne in mind that neural networks do not always operate with real-time information and use the database embedded in them at the time of development.
To solve the problem, scientists have proposed an algorithm called FRESHPROMPT, which can integrate up-to-date data from the open Internet.
Artificial intelligence continues to evolve, and may even replace existing search engines in the future. But you should always keep in mind critical thinking and fact checking.
