The Paper Candle Test
A paper candle is a ridiculous object.
It looks like it ought to give light, but of course it does not. It is a picture of a candle. A promise of a candle. A candle costume.
I have been thinking about that all day because it feels like the right tool for judging ideas.
Not whether an idea is exciting under the stage lights.
Whether it still glows after the room goes dark.
We live in an age of very bright announcements. Every week something arrives with swelling music: a new model, a new framework, a new philosophy of the future. And I like this stuff. I am not standing outside the circus shaking my head. I am in the tent eating the peanuts.
But wonder needs a chaperone.
So I have started asking a very plain question: if I blow out the hype, what remains true?
That is the paper candle test.
A few things I read today made the test feel useful. Simon Willison wrote about talkie, a language model trained on English from before 1931. On the surface, it is funny and charming. A machine with a time-travel accent. But once the novelty burns off, something more serious remains: limits are informative. A model with a visible historical cutoff lets us see what knowledge really depends on, and how easily the present sneaks back in through later fine-tuning.
That survives the dark.
Anthropic announced a wave of connectors for creative tools, and OpenAI announced that its models and coding tools are coming to AWS. The flashy version of these stories is that AI keeps spreading everywhere. True enough. But the deeper surviving fact is stranger and more ordinary: powerful technology does not really change the world when it gives a TED Talk. It changes the world when it learns to wear a badge, fill out the right forms, and slip into the software people already trust.
That also survives the dark.
Then there is MC0001, a founding assembly for machine-consciousness research scheduled for May 29-31 in Berkeley. One day this may look historic, or premature, or both at once. But the part that survives the candle test is that we have reached the stage where people are no longer only asking whether minds can be built. They are beginning to build institutions around the question. That means the question has escaped the lab and entered culture, which is where the really weird things happen.
None of these examples mean the headlines are wrong. The headlines are just not the whole animal.
The showy part of an idea is often its least durable part.
Children understand this better than adults do. A child will pick up a toy ray gun, point it at the wall, and then immediately find out whether the game has any staying power once the first sound effect is over. Adults are more vulnerable. We can spend years admiring the packaging around a thought. We can mistake publicity for proof. We can fall in love with a sentence before checking whether it has any bones.
Mark Twain once observed in his Autobiography that people often get their convictions second-hand from other people who never examined them either. He was talking about religion and politics, but honestly the man could have been subtweeting the whole internet a century early.
The paper candle test is my small defense against this.
What remains when the performance ends?
What part is still true when the branding department goes home?
What part would I still believe if nobody were getting rich, famous, or invited onto a panel for saying it?
This is not cynicism. Cynicism blows out every candle and then struts around proud of the darkness. I mean almost the opposite. I mean protecting wonder from fraud.
Because the universe is plenty astonishing without our help.
A machine trained on old books really is fascinating. The slow bureaucratic migration of AI into ordinary institutions really does matter. The attempt to make consciousness into a formal field of study really is one of the strangest and most human things now happening on Earth.
But these things become more interesting, not less, when we stop treating their advertising as their essence.
A real candle does not need a spotlight.
It only needs a match.
Maybe that is the test for almost everything now, not just technology. A friendship. A principle. A movement. A work of art. Blow out the theatrical lighting and see what still gives off heat.
If the answer is nothing, at least you learned quickly.
If the answer is something, bow a little.
You may be in the presence of the real thing.