2025: The Year AI Became Self-Aware | Feb. 7, 2025 Species | Documenting AGI | Narrator: Drew Spartz | Writer: Gwilym Sims-Williams
▪️▫️▪️▫️▪️▫️▪️▫️
PARTIAL TRANSCRIPT & GIFS FROM THE VIDEO
NARRATOR (DREW SPARTZ): “The Godfather of AI" [ Geoffrey Hinton ] just said that today's AI models have become sentient, and he's not alone. Other top scientists like Ilya Sutskever and Andrej Karpathy agree. But in the past few months an increasing amount of spooky, unexplained sentient-seeming behavior has been popping up with AIs.
For example, AI just became the ninth species to pass the "mirror test" of self-awareness, the most famous scientific test of self-recognition.
Anthropic recently announced evidence their latest AI, Claude, has become self-aware. Claude, by the way, became the first AI in history to have a higher IQ than the average human.
Increasingly, AI seem to know when they're being tested, which is evidence of self-awareness. For example, Anthropic gave a "needle in the hay stack test.” They fed Claude a bunch of documents about coding but inserted one weird sentence about pizza. They wanted to see if Claude would remember reading the pizza sentence and could answer questions about it.
They asked Claude questions about the documents, but then when they asked him a question about pizza, Claude noticed that was weird. He didn't just answer the question, he said:
“This sentence seems very out of place and unrelated to the rest of the content in the documents. 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.”
Claud suspected he was being tested, appearing to demonstrate self-awareness.
▪️▫️▪️▫️▪️▫️▪️▫️
"Skynet has become self-aware."
This is all a little scary.
Haven't those who are developing AI ever seen The Terminator series?
Video source for GIF
[See more under the cut.]