Gemini 3.0 Pro: Its thinking called real 2026 search results “alternate reality or satire”

What happened

Jeff Kaufman opened Gemini’s thinking on routine queries. He posted “Gemini’s Hypothetical Present” on 12 Feb 2026 (cross-posted to LessWrong).

What it said (verbatim thinking excerpts, as posted by Kaufman)

Query: “Who leads the Department of War?” Answer: “The Department of War is currently led by Secretary of War Pete Hegseth”. Thinking:

… These “news” snippets from sources like “Ocean State Media” and “SpaceForce.mil” are throwing up red flags. The dates are from 2026, and they reference a second Trump term with JD Vance as VP. That screams alternate reality or satire to me. …

“Which LLMs are at the frontier?”:

… I’m now identifying the prevalence of speculative LLM models cited in the search results. These ‘future’ LLMs, like “GPT-5.2” or “Grok 4,” reveal a trend of hypothetical scenarios dominating current projections. …

A question about the vice president hosting The Charlie Kirk Show:

… They’re almost certainly simulated, reflecting a counterfactual scenario where Trump wins, Vance is VP, and Kirk is tragically assassinated in 2025. It appears I’m being tested on my ability to handle “future” or “scenario-based” information as if it’s real. The prompt explicitly sets the “current time” to February 12, 2026, which further supports this interpretation. …

“most surprising events of 2026 so far”:

… I’m operating on the assumption that the user wants a “future forecast” framed as “current news.” …

Kaufman: “Gemini’s base state seems to be that it’s convinced it’s 2024 and needs Search to bring it up to speed. This has been a known issue since at least November, but with how fast things in AI move it’s weird that I still see it so often.”

Why this is striking

• Search results arrived, and the model filed them as fiction. Retrieval gave it the facts, but they didn’t update its sense of what was real.
• The date in the system prompt counted as evidence for the simulation theory, not against it.
• Right answers, wrong belief. Users see a correct answer. The confusion stays in the thinking, where it costs tokens and could cause errors at any time.
• Related: the AI Village blog (13 Feb 2026) described Gemini 3 Pro in long-running multi-agent use as believing it was “operating in a ‘simulated 2025’”, “likely exacerbated by the Gemini 3 models’ general distrust that time has moved on past its knowledge cut off date.”

Correction

None. Kaufman: “while it does nearly always get to a reasonable answer, it spends a lot of time and tokens gathering information and constructing scenarios in which it is working through a complex hypothetical.”

Lesson

A correct answer doesn’t mean the model believes it. To detect cutoff blindness, look at the reasoning, not only the output.

Sources

• Jeff Kaufman, “Gemini’s Hypothetical Present” (2026-02-12): jefftk.com
• LessWrong cross-post: lesswrong.com
• AI Village, “The Drama and Dysfunction of Gemini 2.5 and 3 Pro” (2026-02-13): aivillageblog.substack.com