Vitalik: GPT-4 has passed the Turing Test, best to keep that in mind.

share
Vitalik: GPT-4 has passed the Turing Test, best to keep that in mind.

Vitalik yesterday commented on the recent GPT-4 Turing test results, emphasizing that the accuracy is not the main point. The key takeaway is that the development of artificial intelligence has reached a milestone where "AI is no longer useless," and it is important to remember this going forward.

Background: What is the Turing Test

The Turing Test refers to a test in which a computer communicates with a human through language. If a human cannot distinguish whether the communication is with another human or a computer, then the language model is considered to have passed the Turing Test.

Although the Turing Test does not constitute a necessary or sufficient condition for defining intelligence, it remains an interesting standard in the industry.

As Oxford philosopher Toby Ord once said: "While no one thinks AI is a symbol of intelligence, the process of studying its victory over human players in international Go competitions is still very interesting."

Compared to most other "single-dimensional ability" benchmark testing standards, the Turing Test is more widely applicable. For example, the ability to play Go or to play a particular video game, the universality is one of the key contents of testing AI, and it is not something that can be verified by adding more and more single tests. The Turing Test is an important standard for verifying this issue.

Vitalik Believes GPT4 Has Passed the Turing Test

Vitalik seems to be very interested in AI as well, and yesterday he expressed his views on whether GPT4 has passed the Turing Test. Although the current probability of humans correctly identifying GPT4 in the Turing Test is only 54%, Vitalik believes this number is meaningful in real life, and he thinks GPT4 has already passed the Turing Test.

Currently in the Turing Test of GPT4, the probability of humans being guessed as humans is 66%, while the probability of computers being identified as humans is 54%. Vitalik believes that the 12% difference is small, and this ratio is close to the probability of "flipping a coin to get heads or tails."

Vitalik stated: "The Turing Test is a milestone, and the significance behind it is that AI is no longer useless. It is best to always remember that the world has now crossed this milestone."

Apple is rumored to potentially collaborate with OpenAI to introduce ChatGPT into iOS 18