bobcat
Well-known Member
- Location
- Northern Calif
A UC San Diego test was carried out in an online hub with the following:
A group of 300 people were used as participants
In the test each participant was randomly assigned to be either an interrogator, or one of the witnesses being interrogated, and the remaining witness was a chatbot, and they did 8 rounds.
In the first series, the LLM chatbots were only instructed that they were going to be in a Turing Test, and the goal is to convince the interrogator that you are human. In that series Chat GPT 4.5 was only more convincing than the actual human a little over 1/3 of the time.
In the next series the chatbot was instructed to adopt a specific persona, like a young person who's knowledgeable about the internet and culture.
Remarkably in this test, Chat GPT 4.5 was voted more convincing than the human witness 73% of the time. It was voted to be human significantly more than the real human.
Granted, this is only a thought experiment test, and should by no means indicate that an AI is capable of surpassing human mental abilities, but it certainly indicates that specific LLM's are becoming master conversationalist. It's intriguing, and I'm sure this will be getting lots of attention in the ever-evolving perception of technology and the future.
A group of 300 people were used as participants
In the test each participant was randomly assigned to be either an interrogator, or one of the witnesses being interrogated, and the remaining witness was a chatbot, and they did 8 rounds.
In the first series, the LLM chatbots were only instructed that they were going to be in a Turing Test, and the goal is to convince the interrogator that you are human. In that series Chat GPT 4.5 was only more convincing than the actual human a little over 1/3 of the time.
In the next series the chatbot was instructed to adopt a specific persona, like a young person who's knowledgeable about the internet and culture.
Remarkably in this test, Chat GPT 4.5 was voted more convincing than the human witness 73% of the time. It was voted to be human significantly more than the real human.
Granted, this is only a thought experiment test, and should by no means indicate that an AI is capable of surpassing human mental abilities, but it certainly indicates that specific LLM's are becoming master conversationalist. It's intriguing, and I'm sure this will be getting lots of attention in the ever-evolving perception of technology and the future.
