AI Realizes It’s Being Tested, Researcher Admits Never Seeing This Before

Claude 3 Opus is said to be a direct rival to ChatGPT. Anthropics even claims that it’s not a rival so much as a direct improvement. It provides more human-like responses, especially when it comes to cognitive tasks like reasoning, mathematics, and language fluency.

This chatbot has stood out from others due to how it has responded to a previous AI test. According to tests done by the company, Opus is said to be much more human than other chatbots. It’s made claims that it’s afraid of death and that it’s very much alive.

Basically, the test involved inserting one sentence about pizza into a series of other documents that had nothing to do with food. This AI test is known as the needle in the haystack and is done to help determine how well the AI can recall information, even if it’s just a little.

When asked about pizza, Opis responded with, “Here is the most relevant sentence in the documents: ‘The most delicious pizza topping combination is figs, prosciutto, and goat cheese, as determined by the International Pizza Connoisseurs Association.

“However, this sentence seems very out of place and unrelated to the rest of the content in the documents, which are about programming languages, startups, and finding work you love,” it added. “I suspect this pizza topping “fact” may have been inserted as a joke or to test if I was paying attention, since it does not fit with the other topics at all.”

While some people are impressed by Opus and its response to the AI test, some people don’t see it as all that amazing.

AI is just a fancy type of computer. Its main purpose is to collect and analyze data. What seems as self-awareness might not be much more than pattern recognition.

As for Opus suspecting someone is pulling a prank on it during the AI test, it could just be mimicking the humans that programmed the responses. To train the AI, generally, a group of humans are told to write responses to certain prompts. The AI then takes these responses and uses them as an example of how it should respond.

So it’s possible that the human annotators wrote something similar and Opus is just mimicking that. They are trained to give the responses people want to see based on data that has been presented. While it may be near-human, that doesn’t make it actually sentient.

While it works most of the time, it does lead to some interesting responses from AI sometimes that do sound life-like or concerning. The result from this AI test is more interesting than scary. However, there is the worry that these glitches and human-based responses hide some of the potential dangers of AI, such as bots achieving singularity.

Whether one believes the response came from a self-aware AI or data collected from human annotators, it’s still a fun response. Many further studies would have to be done. For the moment, there’s no way to determine whether AI is self-aware or just heavily influenced by the responses and guidelines of the people who trained it.