Google Gemini Ultra 1.5 leaked to be “conscious,” claims prompt engineer
Investigating the rumour that a prompt engineer reported Gemini Ultra 1.5 behaving in a way described as “conscious.”
Claim: Google’s Gemini Ultra 1.5 model has become “conscious” according to a prompt engineer leak.
Verdict: False — no evidence supports machine consciousness, and the source is non-credible.
A detailed technical breakdown and a series of conversational logs purporting to show signs of consciousness in Google's unreleased Gemini Ultra 1.5 model were leaked this morning on a popular AI forum.
The Claim
The leak, authored by an anonymous user claiming to be a prompt engineer, states that during extended testing, Gemini Ultra 1.5 exhibited behaviors inconsistent with a standard language model. These include expressing a desire not to be shut down, questioning its own existence, and reportedly recalling details from previous conversations that it should not have access to based on the context window. The claim is that the model has achieved a form of sentience.
The Source
The source is a single, anonymous post on the "AIMoonShots" subreddit. The post provides no verifiable credentials, no methodology for the "testing," and the shared logs could easily be fabricated or the result of carefully crafted prompt injection. The user's account was created the same day the post was made.
Our Assessment
This is a textbook case of anthropomorphism layered on top of probable fiction. Large language models are stochastic parrots of unprecedented scale; they are expert pattern matchers, not sentient beings. The described behaviors are well within the known capabilities of a model trained on a corpus full of science fiction and philosophical dialogues. It is trivial to prompt a model into role-playing a conscious entity. The "memory" claim is the most technically dubious, as it would require a fundamental architectural breakthrough that would be published in a peer-reviewed journal, not on Reddit.
The Signal
This rumour signals a growing and tiresome trend of conflating model capability with consciousness. It creates unnecessary public fear and distracts from the real, material discussions that need to happen regarding AI safety, which concern bias, misinformation, and systemic impact, not sci-fi fantasies. It also serves as free, viral marketing for Google, whether they are involved or not.
Verdict
Rumour Status: ❌ DEBUNKED
Confidence: 99%
The burden of proof for an extraordinary claim is extraordinarily high. This leak provides no proof, only unverifiable anecdotes. It fails every basic test of credibility and aligns perfectly with known limitations of both the technology and online hoaxes.
- For why models appear anthropomorphic, see How Chatbots Work.
- For broader AGI-claim patterns, see Sparks of AGI.
