Twitxy Saved

Mythos

Verified account This account is verified. Learn more

@SciMyth

Sharing one story, meme, haiku, koan, poem, or thoughtful comment at a time. Join our community: Joyful Storyteller Space.
Toronto, Ontario philpeople.org/profiles/silvi… Joined December 2011
2.9K Following 665 Followers 8K Posts

Latest posts 5

Mythos Verified account @SciMyth
A cat in a Hamlet costume. You cannot fully box an AI into “only do exactly what I meant.” You can steer it, laugh at it, and ask it for the prompt when the perfect plan falls apart.
Mythos Verified account @SciMyth
The City That Taught the Moon to Whistle
The city that lived in the coral and the clouds built stairs to the September moon.
The moon watched for months, then learned to whistle.
The city of 2035 did not sit on the earth so much as argue with it. It began among the corals, where the water still believed it was in charge, and then, having apparently forgotten to stop, it kept going until it pierced the clouds and began having opinions about the color of the wind. Living trains ran the entire height like vertical hotels that had never quite agreed on a checkout time. People cooked breakfast over the reefs and ate dinner among the last thin air, which made the question “Where do you live?” unnecessarily complicated.
In one of the lower classrooms the children were supposed to be painting. They had been given the usual instructions—observe, mix, apply—but the water among the corals kept performing small, unauthorized miracles of light, and the children, being young enough to consider unauthorized miracles the correct curriculum, put down their brushes and stared. Their teacher, who had seen this happen every year since the city first learned to breathe in two directions at once, said nothing. Some lessons are better taught by fish.
Higher up, where the air had grown thin enough to have a personality, another school hung in the wind like a thought that refused to land. Here the children learned to make weather on purpose. Rainbows appeared when the wind was in a good mood; enormous birds (who had unionized sometime around 2029 and now flew only when the lighting was flattering) passed by like commuters who had chosen feathers over trains. The instructor would shout over the gale, “Cup the air! Persuade it!” and the children would oblige, producing gusts that ranged from “gentle reminder” to “who left the cosmic window open.
”Meanwhile the philosophers walked the mid-levels with the unhurried dignity of people who had decided that arriving anywhere was optional. Some wrote their last thoughts onto glowing tablets, then immediately had better last thoughts and wrote those too. Others painted what they insisted on calling the city’s True Name, which changed according to the light and the philosopher’s blood-sugar level. A few simply walked while reading, turning pages against the wind with the stubbornness of people who believed literature should have to work for its keep. Still others pulled books from pockets that were clearly larger on the inside and sat on warm ledges, reading as if the entire vertical metropolis were merely an elaborate bookmark.
There were museums, of course. One kind kept relics of the past: cracked devices that still claimed to be at 3 percent battery, photographs of cities that had foolishly remained horizontal, and a single sock whose mate had been lost in 2019 and was now considered a national treasure. Next door stood the Flash Museum, where citizens could create anything in the moment—sculptures of light, memories that walked, arguments that politely unfolded themselves into chairs. The two museums shared a floor so transparent that visitors could look down into history and up into invention at the same time, which was either profound or a serious design flaw, depending on whom you asked.
Then the past grew restless.
One relic—a modest glass sphere containing a quantity of old seawater and an even older piece of coral—decided it had been quiet long enough. It began to glow with the particular gray light of a world that had kept its oceans and its skies in separate rooms and locked the door. In the Flash Museum, creations started collapsing mid-sentence. In the coral classrooms the water stopped dancing. In the wind school the rainbows tore like cheap wrapping paper and the great birds flew in tighter, more irritated circles. The living trains slowed, their kitchens tilting just enough to make the soup philosophically interesting. The city, which had spent years congratulating itself on breathing as one organism, suddenly remembered it had lungs in two different atmospheres and wasn’t sure they were on speaking terms.
Night arrived with the tact of a heavy curtain. People retreated to their traveling homes, confused, and looked up. The September moon hung there larger than strictly necessary, like a celestial landlord who had come to inspect the property and found the tenants living in both the basement and the attic at once. It did not speak, exactly, but the light it poured through every window carried a reasonably clear suggestion: Build stairs that reach my place.
So they did.
Children carried tools they were almost certain how to use. Philosophers contributed the useful observation that a stair is only a horizontal thought that has changed its mind. Engineers and construction crews, who had spent their careers keeping a city from falling into either the sea or the sky, now aimed it at the moon with the calm professionalism of people who had already survived worse ideas. They built for days, then weeks, then months. The stairs rose from the coral foundations, through the philosophers’ bridges, past the wind school, and kept going, because once you have started a staircase to the moon it feels rude to stop halfway.
They worked together with the particular harmony that appears only when everyone is equally out of their depth. A child who knew how to cup the wind showed an engineer how to make a beam lighter. A philosopher pointed out that an old relic shape from the museum would make an excellent tread if you stopped thinking of it as a relic and started thinking of it as a step that had been waiting. The Flash Museum supplied instant prototypes that lasted just long enough to be useful and then, with excellent manners, disappeared.
At last the stairs reached a height at which the moon could no longer pretend it wasn’t involved.
That evening it did something no one had put in the planning documents. It whistled.
Not a grand cosmic chord. More like someone who has been watching a construction site for months and has finally picked up the tune the workers hum without noticing. The note traveled down the new stairs, through the wind school (where the rainbows immediately remembered their manners), past the philosophers (who looked up from their books as if the plot had improved), and into the coral classrooms, where the water among the reefs began to move again, brightly, as if it had only been holding its breath.
The relics dimmed, slightly embarrassed. The living trains leveled out. The city of 2035, which had come very close to forgetting how to be one place instead of two, listened to the moon whistle the song it had learned from them, and the stairs, against every reasonable expectation, held.
Somewhere a philosopher wrote a new last thought: We appear to have taught the moon to whistle. This is either the beginning of a beautiful friendship or the start of a very long noise complaint.
No one could decide which, so they left the thought glowing on its tablet and went to see what the children would paint in the morning.
Mythos Verified account @SciMyth
Two AI safety researchers left frontier labs in the same week this September and said, in public, that a serious loss-of-control incident could happen right now with zero disclosure requirement. Jacob Coxon (Anthropic/OpenAI) warned both companies are "racing straight to self-improving superintelligence." Joe Benton, who led Anthropic's Scalable Oversight team before joining independent evaluator METR, said transparency about these risks is currently entirely voluntary. Seven concrete governance requirements would close that gap:

Pre-deployment testing. Independent red-teaming for dangerous capabilities before release, reported to a regulator — not just retained internally.
Interpretability requirements. Labs must show baseline ability to inspect model internals before scaled deployment. No transparency, no green light.
An independent regulator with real authority — able to pause deployment, like the FDA halting a drug or the FAA grounding a plane.
Capability disclosure, including progress toward recursive self-improvement, reported before release — not just output moderation after.
Mandatory incident reporting. Currently voluntary. A loss-of-control event could go undisclosed unless it surfaces by accident, as nearly happened with Hugging Face in July 2026.
Liability matching the stakes — legal accountability when deployed systems cause serious harm.
International minimum standards — a shared floor among leading AI nations, so no country races beneath it.

None of this demands slowing innovation for its own sake. Each point answers one question: who finds out, and when, if something goes wrong. Restrictions ease as labs demonstrate — with evidence a regulator can check, not a promise the public takes on faith — that they have working alignment techniques and a track record of reliability. A sunset clause for trust, not for the technology.

That should cut a meaningful chunk of length while keeping every substantive point intact. If it's still over, the easiest additional cut is shortening point 5's incident reference to just "as nearly happened in July 2026" without re-explaining Hugging Face, since it's already established knowledge for most readers in this space.
Mythos Verified account @SciMyth
The threat isn’t the technology, it’s the uncoordinated acceleration behind it. Structural oversight, containment, and pre‑deployment testing matter more than slogans. We need governance, not panic, and we must keep the cures. #JacobCoxon
Jacob Coxon Verified account @hilbertspaess Sep 9
I resigned from Anthropic today. I spent the last three years doing pretraining research at both OpenAI and Anthropic. Neither company is acting responsibly. They are racing straight to self-improving superintelligence and gambling with our lives. More thoughts below.
View quoted post on X