@ejc3 - Ejc3 Mocks Dennison Bertram's
@ejc3 on X Eric Cabrera (ejc3) replies to Dennison Bertram, saying the puzzle's "3811" is a 30x30 grid rather than 12x12, so the instructions fail to add up. He jokes that the model's output only sounds convincing, which is why it "hallucinates better."
- The grid size is wrong: 3811 is 30x30, not the 12x12 the instructions assume.
- The instructions as written don't make sense against that grid.
- The output looks plausible but is wrong, a classic case of confident model hallucination.
The post is a short reply from @ejc3 to a tweet by @DennisonBertram, and it's a pointed correction about a model's output. Bertram had apparently asked a language model about the number 3811, and the model responded with a claim that 3811 is 12x12. @ejc3 shoots that down: 3811 is 30x30, which is 900, not 144. The complaint isn't just that the arithmetic was wrong, but that the model's explanation for it was built on the wrong premise, so none of its supporting steps hold together.
What makes the critique sharp is the closing line: the model "just hallucinates better." That's a jab at the idea that more fluent, confident answers are a sign of reliability. A wrong answer delivered smoothly, with plausible-sounding intermediate steps, can be more misleading than a clumsy one, because the errors blend into the reasoning and are harder to spot. The point is less about one bad calculation and more about how these systems can fabricate a coherent-looking chain of logic around a false starting fact.
For someone skimming their read-it-later queue, the takeaway is brief: a short, sardonic check on an AI arithmetic failure, worth reading mainly as a reminder to verify numerical and factual claims rather than trusting the polish of the explanation. There's no linked source or data beyond the two numbers and the exchange itself.